Governance in the Age of Generative AI: A 360º Approach for Resilient Policy and Regulation

Governance in the
Age of Generative AI:
A 360º Approach for Resilient
Policy and Regulation
W H I T E P A P E R
O C T O B E R 2 0 2 4
In collaboration
with Accenture

Images: Getty Images, Midjourney
Disclaimer
This document is published by the
World Economic Forum as a contribution
to a project, insight area or interaction.
The findings, interpretations and
conclusions expressed herein are a result
of a collaborative process facilitated and
endorsed by the World Economic Forum
but whose results do not necessarily
represent the views of the World Economic
Forum, nor the entirety of its Members,
Partners or other stakeholders.
© 2024 World Economic Forum. All rights
reserved. No part of this publication may
be reproduced or transmitted in any form
or by any means, including photocopying
and recording, or by any information
storage and retrieval system.
Contents
Foreword 3
Executive summary 4
Introduction 5
1 Harness past 6
1.1
Examine existing regulations complicated 6
by generative AI attributes
1.2
Resolve tensions between policy objectives 9
of multiple regulatory regimes
1.3
Clarify expectations around 10
responsibility allocation
1.4
Evaluate existing regulatory authority capacity 11
for effective enforcement
2 Build present 12
2.1
Address challenges of stakeholder groups 12
2.2
Facilitate multistakeholder knowledge-sharing 18
and interdisciplinary efforts
3 Plan future 21
3.1
Targeted investments and upskilling 21
3.2
Horizon scanning 22
3.3
Strategic foresight 25
3.4
Impact assessments and agile regulations 25
3.5
International cooperation 26
Conclusion 27
Contributors 28
Endnotes 33
Governance in the Age of Generative AI 2

Foreword
We are living in a time of rapid innovation and global
uncertainty, in which generative artificial intelligence
(AI) stands out as a transformative force. This
technology impacts various industries, economies
and societies worldwide. With the European Union’s
(EU’s) AI Act now in effect, we have a precedent
for comprehensive AI regulation. The US, Canada,
Brazil, the African Union, Japan and China are also
developing their own regulatory approaches. This
pivotal moment calls for visionary leadership and a
collaborative approach to anticipatory governance.
Over the past year, the AI Governance Alliance
has united industry and government with civil
society and academia, establishing a global
multistakeholder effort to ensure AI serves the
greater good while maintaining responsibility,
inclusivity and accountability. We have been able
to position ourselves as a sounding board for
policy-makers who are grappling with the difficulties
of developing AI regulatory frameworks, and to
convene all players from the AI value chain to
create a meaningful dialogue on emerging AI
development issues.
With Accenture as its knowledge partner, the
Alliance’s Resilient Governance and Regulation
working group (composed of over 110 members),
has contributed to shaping a shared understanding
of the global regulatory landscape. The group has
worked to establish a comprehensive governance
framework that could be used to regulate
generative AI use well into the future.
This paper is a culmination of those efforts and
equips policy-makers and regulators with a clear
roadmap for addressing the complexities of
generative AI by examining existing regulatory
gaps, the unique governance challenges of
various stakeholders and the evolving forms of this
technology. The outputs of this paper are designed
to be practical and implementable, providing global
policy-makers with the tools they need to enhance
generative AI governance within their jurisdictions.
Through this paper, our AI Governance Alliance:
Briefing Paper Series, launched in January 2024,
and our events and community meetings, we seek to
create a tangible impact in AI literacy and knowledge
dissemination.
Given the international context in which this
technology operates, we advocate for a harmonized
approach to generative AI governance that
facilitates cooperation and interoperability. Such
an approach is essential for addressing the global
challenges posed by generative AI and for ensuring
that its benefits are shared equitably, particularly
with low-resource economies that stand to gain
significantly from its responsible deployment.
We invite policy-makers, industry leaders,
academics and civil society to join us in this
endeavour. Together, we can shape a future where
generative AI contributes positively to our world and
ensures a prosperous, inclusive and sustainable
future for all.
Arnab Chakraborty
Chief Responsible
AI Officer, Accenture
Cathy Li
Head, AI, Data and Metaverse;
Deputy Head, Centre for the
Fourth Industrial Revolution;
Member, Executive Committee,
World Economic Forum
Governance in the Age of Generative AI:
A 360º Approach for Resilient Policy and Regulation
October 2024

Executive summary
The rapid evolution and swift adoption of generative
AI have prompted governments to keep pace and
prepare for future developments and impacts.
Policy-makers are considering how generative
artificial intelligence (AI) can be used in the
public interest, balancing economic and social
opportunities while mitigating risks. To achieve this
purpose, this paper provides a comprehensive
360° governance framework:
1
Harness past: Use existing regulations and
address gaps introduced by generative AI.
The effectiveness of national strategies for
promoting AI innovation and responsible
practices depends on the timely assessment
of the regulatory levers at hand to tackle the
unique challenges and opportunities presented
by the technology. Prior to developing new AI
regulations or authorities, governments should:
– Assess existing regulations for tensions
and gaps caused by generative AI,
coordinating across the policy objectives
of multiple regulatory instruments
– Clarify responsibility allocation through
legal and regulatory precedents and
supplement efforts where gaps are found
– Evaluate existing regulatory authorities
for capacity to tackle generative
AI challenges and consider the
trade-offs for centralizing authority
within a dedicated agency
2
Build present: Cultivate whole-of-society
generative AI governance and cross-sector
knowledge sharing.
Government policy-makers and regulators
cannot independently ensure the resilient
governance of generative AI – additional
stakeholder groups from across industry,
civil society and academia are also needed.
Governments must use a broader set of
governance tools, beyond regulations, to:
– Address challenges unique to each
stakeholder group in contributing to
whole-of-society generative AI governance
– Cultivate multistakeholder
knowledge-sharing and encourage
interdisciplinary thinking
– Lead by example by adopting responsible
AI practices
3
Plan future: Incorporate preparedness and
agility into generative AI governance and
cultivate international cooperation.
Generative AI’s capabilities are evolving
alongside other technologies. Governments
need to develop national strategies that
consider limited resources and global
uncertainties, and that feature foresight
mechanisms to adapt policies and regulations
to technological advancements and emerging
risks. This necessitates the following
key actions:
– Targeted investments for AI upskilling
and recruitment in government
– Horizon scanning of generative
AI innovation and foreseeable risks
associated with emerging capabilities,
convergence with other technologies
and interactions with humans
– Foresight exercises to prepare for
multiple possible futures
– Impact assessment and agile
regulations to prepare for the downstream
effects of existing regulation and for future
AI developments
– International cooperation to align
standards and risk taxonomies and
facilitate the sharing of knowledge
and infrastructure
Governments should address regulatory
gaps, engage multiple stakeholders in
AI governance and prepare for future
generative AI risks.

Introduction
As organizations and individuals consider how best
to adopt generative artificial intelligence (AI), new
powerful capabilities continue to emerge. For some,
humanity’s future with generative AI can feel full of
promise, and for others, concern. Indeed, across
industries and sectors, generative AI presents
both opportunities and risks. For example – will
generative AI enhance personalized treatment
plans improving patients’ health outcomes, or will
it induce novel biosecurity risks? Will journalism be
democratized through new storytelling tools, or will
disinformation be scaled?
There is no single guaranteed future for generative
AI. Rather, how society adapts to the technology
will depend on the decisions humans make in
researching, developing, deploying and exploiting
its capabilities. Policy-makers, through effective
governance, can help to ensure that generative AI
facilitates economic opportunity and fair distribution
of benefits, protects human rights, promotes
greater equity and encourages sustainable
practices. Governance decisions made now
will shape the lives of present and future
generations, how (and whether) this technology
benefits society and who is left behind.
In response to the continued growth of the
generative AI industry and rapid adoption of
its applications across the world, this paper’s
360° framework outlines how to build resilient
governance that facilitates AI innovation while
mitigating risks, from the development stage to
its use. The framework is designed to support
policy-makers and regulators in the development
of holistic and durable generative AI governance.
The specific implementation of the framework,
however, will differ between jurisdictions,
depending on the national AI strategy, maturity of
AI networks, economic and geopolitical contexts,
individuals’ expectations and social norms.
A 360° framework is needed for resilient
generative AI governance, balancing innovation
and risk across diverse jurisdictions.
A 360º approach for resilient policy and regulation
F I G U R E 1
360º
governance
P
i
l
l
a
r
1
:
H
a
r
n
e
ss past Pillar 2
:
B
u
i
l
d
p
r
e
s
e
n
t
Pillar 3: Plan future
Encourage whole-of-society generative
AI governance and cross-sector
knowledge sharing.
Make use of existing regulations and
address gaps caused by generative AI.
Incorporate preparedness and agility
in generative AI governance and
facilitate international cooperation.

Harness past
Greater clarity and certainty regarding
existing regulatory environments is necessary
to address emerging generative AI challenges
and opportunities.
With increasing
digitalization and
a growing trend
of monetizing
personal and
professional data,
protection of
privacy is both
vital and complex.
Policy-makers are
looking to prioritize
privacy-preserving
considerations.
1.1
Examine existing regulations complicated
by generative AI attributes
Successful implementation of national strategies
for responsible and trustworthy governance
of generative AI requires a timely assessment
of existing regulatory capacity – among other
governance tools – to tackle the unique
opportunities and risks posed by the technology. This
includes examination of the adequacy of existing
legal instruments, laws and regulations, resolution
of regulatory tensions and gaps, clarification
of responsibility allocation among generative AI
supply chain actors and evaluation of competent
regulatory authorities’ effectiveness and capacities.
Such assessments must respect the fundamental
rights and freedoms already codified in international
human rights law, such as the protection of
particular groups (e.g. minority rights1
and children’s
rights2
) as well as legal instruments that are domain-
specific (e.g. to cybercrime3
and climate change4
).5
While generative AI’s emerging properties and
capabilities may warrant novel regulations, policy-
makers and regulators should first examine their
jurisdiction’s existing regulations for addressing
new challenges. They should also identify where
existing regulations may be applied, adapted or
foregone to facilitate the objectives of a national
AI strategy. Navigating generative AI’s interactions
with existing regulations requires a nuanced
understanding of both the technical aspects
and the legal principles underlying the impacted
regulations. Table 1 discusses examples of how
regulatory instruments can be complicated in the
context of generative AI.
Privacy and data protection
Generative AI models amplify privacy, safety and
security risks due to their reliance on vast amounts
of training data, powerful inference capability and
susceptibility to unique adversarial attacks that
can undermine digital trust.6
A number of risks
arise from the inclusion of personal, sensitive and
confidential information in training datasets and user
inputs, lack of transparency over the lawful basis for
collecting and processing data, the ability of models
to infer personal data and the potential for models
to memorize and disclose portions of training data.
With increasing digitalization and a growing trend
of monetizing personal and professional data,
protection of privacy is both vital and complex.
Policy-makers are looking to prioritize privacy-
preserving considerations applicable to digital data
while also creating affordances for data pooling that
could lead to AI-facilitated breakthroughs.7
Such
affordances could be made to promote innovation
for public goods in areas such as agriculture,
health and education, or within narrowly specified
exceptions for data consortia that facilitate the
training of AI models to achieve public policy
objectives.8
Another emerging issue for policy-
makers is that of ensuring generative AI safety
and security, even when it may involve interaction
with personal data, as in the case of investigating
and responding to severe incidents. This could
be addressed through the creation of regulatory
exceptions and guardrails to ensure both privacy
and responsible AI outcomes.
Pillar 1

Copyright and
intellectual property
Generative AI raises several issues relating to
copyright infringement, plagiarism and intellectual
property (IP) ownership (see Issue spotlight 1),
some of which are currently being considered
by courts in various jurisdictions. Rights related
to protecting an individual’s likeness, voice and
other personal attributes are also implicated by
the creation of “deepfakes” using generative AI.
A blanket ruling on AI training is uncertain and
judges could determine the fairness of certain data
uses for specific products based on the product’s
features or outputs’ frequency and similarity to
training data.9
Looking ahead, there is a pressing
need for comprehensive examination of regulatory
frameworks and for necessary guidance on
documenting human creativity in the generation
of content as a means of asserting IP protection.
Training generative AI systems on copyright-protected data, and tensions with
the text and data mining exception
I S S U E S P O T L I G H T 1
Text and data mining (TDM) is the automated
process of digitally reproducing and analysing
large quantities of data and information to
identify patterns and discover research insights.
Various jurisdictions around the world – such
as Japan, Singapore, Estonia, Switzerland and
the European Union (EU) – have introduced
specific exemptions within their copyright laws to
enable TDM extraction from copyright-protected
content to innovate, advance science and create
business value.
Given the vast amounts of data that generative
AI systems use to train on and generate new
content, jurisdictions should establish regulatory
clarity regarding TDM for the purpose of
generative AI training. This could be done, for
example, by confirming whether AI development
constitutes “fair dealing” or “fair use” (a key
defence against copyright infringement) or
falls within the exemptions recognized in some
copyright laws. Countries like the UK are exploring
such regulatory exceptions, seeking to promote
a pro-innovation AI agenda.10
Ultimately, there is
mounting pressure on governments to resolve the
copyright tension definitively.11
Licensing and data access on an “opt-in” or
“opt-out” basis are also under examination to
address TDM concerns, in addition to a range of
technologies and standards that attempt to cede
control to creators, allowing them to opt out from
model trainers.12
Licensing proponents argue
that scraping for generative AI training without
paying creators constitutes unlawful copying and
is a form of reducing competition.13
AI developers,
however, argue that requirements to pay
copyright owners for content used in training
would constrain model development, negatively
impact venture capital (VC) funding and reduce
competition among generative AI models.14
While
they do not eliminate IP law concerns entirely,
opt-in/out and licensing efforts could contribute
to setting standards that generative AI foundation
model providers would be expected to uphold.
Consumer protection
and product liability
While AI-specific regulation remains voluntary or
pending in jurisdictions outside of the EU, consumer
regulation and product liability laws continue to
be applicable, regardless of whether they strictly
contemplate AI or other technologies. Generative
AI has the potential to influence the consumer
market by automating various tasks and services.
This may, however, also challenge traditional
approaches to risk assessment and mitigation (due
to the technology’s broad applicability and ability
to continually learn and generate new and unique
content), as well as product safety standards
(for example, in health and physical safety). The
development of standards should be an iterative,
multidisciplinary process that keeps pace with
technological advancements.
7
Governance in the Age of Generative AI

Competition
Market authorities must ensure that the competitive
conditions driving the rapid pace of innovation
continue to benefit consumers. Although existing
competition laws remain applicable, generative AI
raises new concerns related to the concentration of
control over critical components of the technology
and certain partnership arrangements. For example,
generative AI’s capabilities are enhanced with
access to high-performance compute capacities
and certain datasets that may prove critical for
model development. The latter can depend on
access to a vast number of users, contributing to
economies of scale that challenge competition.15
In
response, competition authorities around the globe
are starting to provide guidance on competition
risks and expectations in generative AI markets.16
Competition complexities at each layer of the AI
stack will need to be evaluated as the technology
evolves to enable access and choice across AI
models, including general (e.g. ChatGPT), area-
specific (e.g. models designed for healthcare) and
personal use models. Such evaluations will also
need to be considered alongside existing legislation
relating to national security, freedom of expression,
media and assembly.
Selection of complexities introduced by generative AI for existing regulatory areas
TA B L E 1
Regulatory area
Emerging complexities
(non-exhaustive)
Emerging strategies under consideration
by regulators (non-exhaustive)
Privacy and data
protection
Legal basis for user data being used to train
generative AI models
Enforcement of data-minimization principles17
and opt-in/out
rights by generative AI providers and deployers18
Incidental collection of personal data
by web-crawlers
Clarifying web terms-of-service agreements and encouraging
privacy-enhancing technological measures such as the
detection and redaction of personally identifiable information19
Specifying purpose limitations for data
collection
Guidance for purpose thresholds within domain-specific
regulations, e.g. financial services20
Online safety and protection of vulnerable
groups, especially minors, from harmful outputs
Position statements highlighting expectations for safety
measures and preferences for emerging best practices21
Copyright and IP
Copyright infringement of training data Clear policy positions and accumulation of legal precedents
on the relations between copyright and generative AI22
IP rights and ownership of works generated
by AI
Guidance on assessing the protectable elements
of AI-generated works23
Attribution and fair compensation for artists
and creators
Investments in solutions for attribution and author recognition
such as watermarking and content provenance, along with
privacy and data protection
Extension of generative AI model training
to additional data modalities (e.g. sensory,
biological, motion)
Considerations of new IP challenges and classifications related
to emerging data modalities
Consumer
protection and
product liability
Liability obligations resulting from scope
of multiple applicable regulations
Considerations around whether and in which cases a concern
is covered by the existing regulations
The lack of a specific purpose of the generative
AI model before its implementation complicates
liability arising from defectiveness and fault
Combining the conventional AI fault and defectiveness
criteria with new methods designed for generative AI’s
technical nuances
Efficacy of evidential disclosure requirements Broadening the disclosure requirement to encourage
transparency via explainability, traceability and auditability,
and include systems that are not just classified as high-risk
Competition
Business conduct or agreements that enable
a dominant firm to exclude rivals
Initiating sectoral studies to develop a baseline understanding
of the competitive dynamics of the AI technology stack,
reviewing agreements between industry players and examining
single firm conduct24
Unfair or deceptive practice Issuing guidance on unfair or deceptive practice prohibitions
if it does not exist25
Impact of downstream applications on
competition across several sectors
Stakeholder consultations on how generative AI impacts
competition in important markets, e.g. search engines,
online advertising, cloud computing and semiconductors26

The intersectional nature of generative AI
technologies and the applicability of multiple
regulatory instruments creates a complex
environment where regulatory frameworks often
overlap and conflict due to competing policy
objectives. As technology evolves and becomes
more widely adopted, regulators must address
emerging tensions and mitigate the risk of
undermining legal certainty and respect for
legitimate expectations.
Addressing tensions between
horizontal regulations
Multiple horizontal regulations, which aim to create
broad, industry-agnostic standards, may conflict
when they impose requirements that are difficult
to reconcile across generative AI contexts or
applications. For example, generative AI model
developers may have trouble identifying the
appropriate lawful basis for data processing and
delivery according to data protection rights articulated
through the EU’s General Data Protection Regulation
(GDPR). A similar tension emerges between copyright
law – which protects the rights of creators and
inventors ensuring that they can control and profit
from their creations – and generative AI innovation,
which often uses copyrighted material for training.
Addressing tensions between
horizontal and vertical regulations
Horizontal regulations may also conflict with
vertical regulations tailored to specific sectors. For
instance, financial institutions using generative AI
may encounter challenges balancing horizontal
privacy regulations with financial sector know-
your-client (KYC) procedures. Where data
protection regulations require organizations to
minimize personal data collection linked to a
specific purpose, KYC guidelines require financial
institutions to conduct thorough due diligence on
clients to ensure compliance with anti-money-
laundering laws.
1.2
Resolve tensions between policy objectives
of multiple regulatory regimes
Regulators must
address emerging
tensions and
mitigate the risk
of undermining
legal certainty
and respect
for legitimate
expectations.

Challenges and considerations for generative AI responsibility allocation (non-exhaustive)
TA B L E 2
Example challenges Considerations for policy-makers
Variability
– Model variations include features (e.g. size), scope
(e.g. use purpose), and method of development
(e.g. open-to-closed source).
– Technical approaches to layering and fine-tuning
continuously evolve, enabling general-purpose
models to adapt functionality for specific applications.
– Entity categorization complexities involve multiple
actors from different sectors with overlapping or
multiple roles.
– Case-based review: Policy-makers should
provide general allocation guidance to cultivate
predictability, but include mechanisms that allow
case complexities to determine precise allocation.
Requiring actors to identify responsibility hand-offs
is one approach being examined by jurisdictions.
– Terminology: Policy-makers should collaborate
to arrive at shared terms for models, applications
and roles, e.g. in line with ISO 420001 from the
International Organization for Standardization (ISO).
– Regulatory carve-outs: Policy-makers should limit
instances when use can lead to unfair advantages,
such as where some entities are able to bypass
crucial safeguards and accountability measures or
engage in regulatory arbitrage.
Disparity
between actors
– Single points of failures and power concentration
occur as a result of a few foundational models
(serving many applications and billions of end users).
– Disparities in influence emerge between upstream
and downstream actors.
– There is limited transparency for downstream actors
related to training data and for upstream actors
related to end-user activity.
– Proportionality: Policy-makers should consider the
control, influence and resources each actor has in
the generative AI life cycle, and ability to redress
issues resulting in harm.
– Third-party certifications: Policy-makers should
consider appropriateness and necessity of using
third parties for a robust AI certification system
(potentially defined through regulation) that enables
actors to verify and trust each other’s capabilities.
Complexity
of review
– Interpretability difficulties relating to outputs arise
due to models often operating as “black boxes” to
varying degrees.
– Traceability difficulties transpire in 1) diversity of data
sources, 2) sequence of events that led to a fault,
3) determining whose negligence or malice induced
the fault or made the fault more likely.
– Physical inspection or verification of changes
to generative AI products in the market has
limited feasibility.
– Documentation: Policy-makers should incentivize
appropriate transparency and vulnerabilities
disclosure upstream and downstream to enable
responsible decisions. Concerns about trade secrets
or data privacy compromise the need to be
mitigated.
– Traceability mechanisms: Policy-makers should
require the ability to trace outputs back to their
origins while considering compromise and mitigation
measures for IP and data privacy concerns.
– Continuous compliance: Policy-makers should
integrate standards for market entry and procedures
for post-approval changes, and encourage industry
review boards and ongoing independent audits.28
As defined in the World Economic Forum’s Digital
Trust Framework,27
maintaining accountability
and oversight for trustworthy digital technologies
requires clearly assigned and well-defined legal
responsibilities alongside remedy provisions for
upholding individual and social expectations.
Generative AI introduces complexities into
traditional responsibility allocation practices, as
examined in Table 2. Policy-makers should consider
where supplementary efforts are needed to address
gaps and where legal and regulatory precedents
can help to clarify generative AI responsibility.
The issuance of effective guidance requires
consideration of how liability within the generative AI
supply chain can vary for different roles and actors
as well as consideration of retroactive liabilities
and dispute-resolution provisions. Unresolved
ambiguity in responsibility allocation can limit
investor confidence, create an uneven playing
field for various supply chain actors and leave
risks unaddressed and harms without redress.
1.3
Clarify expectations around
responsibility allocation

Effective regulatory enforcement depends on
governments identifying the appropriate authority
or authorities and enabling their activity with
adequate resources.
Expansion of existing regulatory
authority competencies
While generative AI may elicit consideration
of a new AI-focused authority, governments
should first assess opportunities to make use of
existing regulatory authorities with unique domain
knowledge and ensure they can translate high-
level AI principles to sector-specific applications.
Considerations of how to delegate regulatory
authority for AI will depend on a jurisdiction’s AI
strategy, resources and existing authorities. For
example, countries that have a data protection
authority (DPA), such as France, tend to rely on the
DPA to comprehensively address AI, since data is
fundamental to AI models and uses. In the same
vein, countries without DPAs, such as the US, may
lack a readily apparent existing authority.
Furthermore, the specific mandate and procedural
frameworks of existing authorities such as DPAs
impact AI governance. For example, Singapore’s
DPA, the Personal Data Protection Commission
(PDPC), sits within a broader authority, the
Infocomm Media Development Authority (IMDA),
whose mission includes cultivating public trust
alongside economic development. Thus, AI
governance from Singapore’s DPA actively
considers both trust and innovation within its
regulations. This underscores how generative AI
may necessitate the expansion of remits for existing
regulators. For example, Singapore’s IMDA must
now consider issues related to generative AI data
ownership and provenance, and the use of data for
model training, including potential compensation for
creators whose content was trained on.
Coordination of
regulatory authorities
Coordination between regulatory authorities
can prevent duplication of efforts and enhance
operational resilience for overburdened and
under-resourced offices. New coordination
roles or responsibilities should be considered.
For example, the UK has created the Digital
Regulation Cooperation Forum (DRCF),
encompassing the Competition and Markets
Authority (CMA), Financial Conduct Authority
(FCA), Information Commissioner’s Office (ICO)
and Office of Communications (Ofcom) to ensure
greater cooperation between regulators on online
matters, including within the context of AI. Similarly,
Australia’s Digital Platform Regulators Forum (DP-
REG) – an information-sharing and collaboration
initiative between independent regulators –
considers how competition, consumer protection,
privacy, online safety and data issues intersect.
Dedicated AI agency versus
distributed authority between
sector-specific regulators
The founding of an AI agency requires careful
consideration regarding, for instance, the scope
of responsibilities, availability of resources and
domain-specific regulatory expertise. For example,
would the agency serve to coordinate, advise and
upskill sector-specific regulators on AI matters,
likely requiring less funding, or would it serve as an
AI regulatory authority with enforcement powers,
requiring greater funding? Some argue that a
central AI agency is needed to address highly
capable AI foundation models.29
Others consider
a central AI agency more prone to regulatory
capture and less effective for AI’s diverse use cases
than distributed regulations among existing sector-
specific authorities with domain-specific knowledge.
Consequently, many would prefer a council-like AI
body that coordinates and advises existing sector-
specific authorities.30
Jurisdictions are finding creative ways to navigate
limited funding and political compromise. For
example, the EU embedded its new AI Office within
the EU Commission,31
instead of setting it up as
a solitary institution, to amplify the effectiveness
of the office’s limited number of staff. Like the EU,
jurisdictions are navigating complex challenges
of how to creatively resource a new AI body or
authority while ensuring its independence. Still,
enforcement of the AI Act, like GDPR, may strain
authorities at the member-state level. For instance,
while Spain has set up a centralized authority
to enforce the act’s provisions, France may use
existing regulators, such as the DPA, as the
authority of record.
Some argue
that a central
AI agency is
needed to address
highly capable AI
foundation models.
Others consider
a central AI agency
more prone to
regulatory capture
and less effective
for AI’s diverse
use cases.
1.4
Evaluate existing regulatory authority capacity
for effective enforcement

Build present
Governments should address diverse
stakeholder challenges to facilitate
whole-of-society governance of generative
AI and cross-sector knowledge sharing.
2.1
Address challenges of stakeholder groups
While regulators play a critical role, they cannot
independently ensure the resilient governance of
a technology that has simultaneously broad and
diversified impacts, and capabilities that continue
to evolve. Other stakeholder groups hold key puzzle
pieces to assembling resilient governance and a
responsible AI system, for example:
– Industry: With proximity to the technology, its
developers and users, industry is at the front
line of ensuring that generative AI is responsibly
governed across countless use cases within
commercial applications and public services.
– Civil society organizations (CSOs): With
expertise on how generative AI uniquely impacts
the different communities and issue spaces they
represent, CSOs enable informed and holistic
policy-making.
– Academia: Through rigorous and independent
research and educational initiatives, academia
is critical to shaping responsible AI development
and deployment and ensuring public literacy on
responsible use.
Governments must use a broader set of
governance tools, beyond regulations, to:
– Address the unique challenges of each
stakeholder group in contributing to society-
wide generative AI governance
– Facilitate multistakeholder knowledge-sharing
and encourage interdisciplinary thinking
– Lead by example by adopting responsible
AI practices
Enable responsible AI
implementation by industry
Governments are carefully considering how to avoid
over- and under-regulation to cultivate a thriving
and responsible AI network, where AI developed
for economic purposes includes robust risk
management, and AI research and development
(RD) is harnessed to address critical social and
environmental challenges. Since market-driven
objectives may not always align with public
interest outcomes, governments can encourage
robust and sustained responsible AI practices
through a combination of financial mechanisms
and resources, clarified policies and regulations,
and interventions tailored to industry complexity.
Incentivize proactive, responsible
AI adoption by the private sector
Public policy-making processes often lack the
private sector’s ability to adopt governance
protocols for innovative technologies. To
address this, governments should assess
the applicability of existing AI governance
frameworks (e.g. Presidio AI Framework,32
NIST AI Risk Management Framework33
) and
encourage proactive industry adoption. In
addition to educating industry on frameworks,
governments can cultivate an environment where
industry is incentivized to proactively invest in
responsible AI. Potential strategies include:
– Financial incentives: Governments could
introduce inducements for responsible AI
practices such as tax incentives, grants or
subsidies for RD, talent or training. Policy-
makers could consider potential tax rate
adjustments to incentivize AI designed
to augment (rather than replace) human
labour,34
and carefully consider trade-offs
of proposed adjustments.
– Sustained funding: Government leaders should
ensure investment in both short- and long-
term RD to reach breakthroughs on complex
AI innovations and address responsible AI
challenges. Jurisdictions with a less advanced
AI industry may require greater initial government
investment to incentivize VC funding.
Pillar 2
Governments
are carefully
considering how
to avoid over- and
under-regulation to
cultivate a thriving
and responsible
AI network, where
AI is harnessed
to address
critical social and
environmental
challenges.

– Procurement power: Governments should
explore preferred procurement measures for
AI with demonstrable responsible AI metrics.
– Access: Governments should provide
opportunities for public-private partnerships and
access to public datasets for AI developed with
demonstrable responsible AI metrics or that’s
designed for social or environmental benefit.
– Responsible AI RD and training: Leaders
should examine the suitability of requiring a
percentage of RD expenditure for responsible
AI governance and/or training for organizations.
Clarify policies and
enable measurement
A responsible AI system is of strategic importance
to investors for mitigating regulatory and non-
regulatory risks (e.g. cyberattacks), and improving
top- and bottom-line growth.35
Over the last
decade, investors have helped drive industry
investment in environmental issues, and they
can play a similar role in incentivizing responsible
AI practices, for instance by addressing AI’s
vast energy use.36
However, uncertainty in how
government AI policies will be implemented and
enforced prevents confident investing in responsible
AI practices.
Governments should set clear national priorities
and policies on responsible AI, reduce ambiguity
in existing regulations and provide signals on the
trajectory of regulations. Singapore’s PDPC, for
example, proactively shared advisory guidelines37
that clarify the application of existing data laws to
AI recommendation and decision systems. The
guidelines additionally highlight exceptions, with
the aim of helping industry navigate regulation.
Encourage businesses to
test, evaluate and implement
transparency measures,
including through:
– Clear frameworks: to measure risks
as well as social and human rights and
environmental impacts
– Certifications: to clarify responsible AI
practices and testing are satisfactory and
draw investors and the public
– Sandboxes: to experiment and refine before
wider deployment, with incentivized participation
– Knowledge-sharing: to promote sharing of
benchmarking, e.g. Stanford AI Index Report38
– Competitions: to address complex AI
challenges, e.g. National Institute of Standards
and Technology (NIST) generative AI challenge39
– Technical standards: to establish common
methodologies and benchmarks for evaluating
AI system performance, safety and ethical
compliance across different domains and
applications. e.g. ISO 4200140
A responsible
AI system is
of strategic
importance
to investors
for mitigating
regulatory and non-
regulatory risks
(e.g. cyberattacks),
and improving
top- and bottom-
line growth.
13

Governance challenges by business size (non-exhaustive)
TA B L E 3
Tailor interventions to diverse
industry needs
Policy-makers need to consider the diversity
of AI governance challenges faced by industry
stakeholders to identify meaningful points of
intervention. Table 3 illustrates how business
size can determine the resources available to
implement responsible AI governance and the
compliance complexities encountered. Other
governance challenges can result from industry
stakeholder characteristics, such as sector,
location, industry maturity, risk sensitivity and
role in the AI supply chain.
Challenges Considerations for policy-makers
Large businesses
Implementation: Difficulties may occur for AI
governance operationalization and compliance within
complex or differently structured organizations.
Policy-makers should provide implementation guidance
that builds upon current risk management frameworks,
global standards, benchmarks and baselines.
Competition: Competitors may not invest equally
in responsible AI practices.
Policy-makers could review responsible AI practices
and regulatory compliance across stakeholders.
Clarity: Navigating compliance ambiguities
or complexities across sectors and between
jurisdictions may present challenges.
Where possible, policy-makers can provide guidance
on what actions are within or outside regulations,
reduce overlap and facilitate interoperability
through harmonization.
Small- and
medium-sized
enterprises
(SMEs) and
start-ups
Resources: Resources to develop and demonstrate
robust responsible AI practices to regulators, investors
or partners may be limited.
Policy-makers should provide guidance, training and
consultation access on AI governance, facilitate insight-
sharing between large businesses and SMEs, and use
certification mechanisms.
Applicability: AI governance frameworks and
recommendations lack applicability or specificity
to the realities of SME operations.
Policy-makers should include input from diverse
SMEs in development of national and international
governance frameworks.
Prioritization: Fast pace of start-ups and lack of capital
can lead to prioritizing innovation over risk assessment.
Policy-makers can incorporate responsible AI practices
and regulatory landscapes into curricula of start-up
accelerators and incentivize participation in sandboxes.

Challenges and considerations for policy-makers to support academic stakeholder
groups (non-exhaustive)
TA B L E 4
For academic institutions For researches For educators
Appropriate use
Clarify compliance with evolving
relevant regulations (e.g. AI, data
privacy and copyright) and simplify
regulations to enable research in
responsible AI.
Provide guidance on responsible
generative AI use in research42
(e.g. data analysis) and training
on risks to boost cognizance
when conducting research in
the age of generative AI (e.g. of
potential misuse by respondents to
online studies).
Provide guidance on responsible
generative AI use by teachers (e.g.
essay review and feedback) and
students (e.g. critical evaluation
of generative AI outputs in essay
writing).
Resources
Ensure access to physical and
digital infrastructure needed for
faculty, researchers and students
to become familiar with AI and use
it responsibly.43
Provide access to data and
compute capabilities to conduct
leading AI and generative AI
research, and clarify guidelines for
accessing public sector data
while maintaining privacy.
Provide regularly updated
training materials and ensure
that educators, regardless of
institutional prestige, can keep
pace with AI advancements.
Funding
Close pay gaps between industry
and academia to reduce AI brain
drain to industry.
Allocate research grants into
responsible AI challenges (e.g.
hallucinations, bias) that do not
require cut-throat competition
or complex applications.
Allocate funding for courses
on AI and responsible AI.
Enable leading AI research
and education by academia
Through research and education, academia is
a critical stakeholder in cultivating a robust AI
network. Until the early 2000s, leading AI RD was
primarily conducted within academia. It contributed
to providing open-source knowledge that
accelerated innovation and optimized development
costs. With recognition of the economic potential of
AI, investment has since shifted RD to industry.
Without academia at the forefront of AI RD,
key risks emerge:
– Homogenization of the AI network
– Decline in discoveries that emerge from
academia’s interdisciplinary research settings
– Decreased independent research around
AI ethics, safety and oversight
– Diminished general workforce training
– Barriers to cross-institution collaboration
– Reduced ability to wield academic freedom
to challenge prevailing consensus
– Broken AI talent pipeline
Since generative AI has extensive and costly
infrastructural needs (e.g. compute capabilities,
data), academia’s ability to conduct leading
research is severely limited.41
Table 4 outlines the
range of challenges facing academic stakeholders
that policy-makers should address to cultivate
a thriving AI system. These challenges must be
considered in the context of the different operating
conditions of academic institutions, for example of
private, public and community colleges, to ensure
equitable access to AI literacy, benefit from the AI
economy and a diverse pipeline of responsible AI
experts. Similarly, policy-makers should address
the unique literacy and access challenges in earlier
educational settings, for example in primary and
secondary schools.

As the largest digital user group and fastest
adopters of technology, children and youth are
at the forefront of AI-enabled systems. The effects
of using generative AI, both positive and negative,
will have wide-ranging and lifelong impacts that
will shape the development, safety and worldviews
of children.44
Research agendas are beginning
to emerge to aid precise policies around the
disproportionate effect of algorithmic bias on
minoritized or marginalized children.45
They can
additionally inform policies that address concerns
around how generative AI training46
and use47
could
amplify child sexual abuse material (CSAM),48
and
how generative AI applications, especially the use
of chatbots and smart toys, may affect cognitive
functioning among children.49
Existing resources
such as UNICEF’s Policy Guidance on AI for
Children,50
the European Commission’s guidance
on Artificial Intelligence and the Rights of the Child51
and the World Economic Forum’s AI for Children
Toolkit,52
provide valuable direction.
Given their limited political agency, economic
influence and organizing power, children can
often be overlooked in technology governance
considerations, even as they are most impacted.
Further, the existence of inequalities around the
digital divide exacerbates the risks and harmful
effects of generative AI for some children more than
others, given their inability to participate in shaping
generative AI’s development or access its benefits.
Engaging young users, their guardians and local
communities in a meaningful and ongoing way
throughout the life cycle of generative AI projects
and governance, directly and via CSOs with deep
technical or policy expertise in these areas, is vital
for children’s empowerment and the development
of responsible AI innovation. Transparency in how
children’s rights and input have been considered
and implemented is critical to promoting public
trust and accountability.53
A child-centric approach for generative AI governance
I S S U E S P O T L I G H T 2
CSOs face significant access and participation
challenges preventing them from assessing
societal impacts of generative AI technologies,
informing governance policies and supply chain
accountability, and advocating for the rights
of citizen groups and vulnerable populations
such as children, as examined in Table 5.
Ensure access and participation
of CSOs
In addition to ensuring technical expertise in
governance conversations, there is a critical need
for expertise related to the social impacts of AI and
generative AI, informed by the lived experiences
of those interacting with the technology. CSOs
play a key role in representing various citizen
groups, individuals and issue spaces and provide
related technical and societal expertise. CSOs
can also offer independent oversight, holding
governments and companies accountable for their
AI implementation.
Depending on their missions, CSOs have unique
expertise around generative AI implications that
policy-makers should make use of, for example:
– Labour protection groups can help inform the
skills and training needed to ensure generative
AI leads to job growth rather than displacement.
– Environmental groups can provide guidance
on ways AI can help address local and global
climate challenges, and considerations
regarding generative AI’s vast energy
consumption.
– CSOs focused on creative practice, journalism,
mis/disinformation or election monitoring can
inform the harnessing of generative AI’s creative
potential while preserving information integrity
and ownership rights.
– CSOs serving marginalized populations or
protected classes can help ensure AI policies
and technologies holistically consider the
varied opportunities and risks posed (see Issue
spotlight 2).

Challenges and considerations for policy-makers to support CSOs (non-exhaustive)
TA B L E 5
Access
Under-resourced: There is a lack of adequate tools
and skills to review impacts of generative AI.
– Policy-makers should provide access and training
for cutting-edge tools and incentivize industry to
share tools.
– They should fund RD to improve tools’
abilities, (e.g. detection in minority languages
or compressed media).
– They should provide funding for CSOs to undertake
independent impact assessments.
Opaque: There are limited metrics on how companies
have implemented responsible AI, including principles
that have been publicly committed to.54
– Policy-makers should standardize and incentivize
responsible AI reporting.
– They should provide CSOs with easier access to
mandated transparency data, (e.g. via EU Digital
Services Act and AI Act).
Limited information: There is a lack of access to
training data and weights, and information on how
companies moderate public use of AI technologies.
– Policy-makers should incentivize industry to share
data with CSOs, while preserving privacy and IP.
– They should standardize transparency reporting
on how AI companies moderate technology use.
Participation
Disempowered: CSO inclusion is often limited
in numbers and in influence on decision-making.
There is even less inclusion of CSOs operating
outside regulatory regimes, which will be
impacted by generative AI and regulatory shifts.
– Policy-makers should ensure sectoral parity
in discussions.
– They should educate on the value of CSO
community-driven insights.
– They should strengthen outreach to vulnerable
communities and relevant CSOs, including
transnational CSOs, and engage international
CSO forums, (e.g. C7, C20, African Union Civil
Society Division).
Delayed: CSOs engaged late in technical
and governance processes.
– Policy-makers should ensure task forces, institutes
etc. have CSO participation at formation.

2.2
Facilitate multistakeholder knowledge-sharing
and interdisciplinary efforts
Governments should facilitate knowledge-
sharing across stakeholder groups and with other
governments to reduce duplicative efforts, offset
expertise gaps and enable informed policies
capable of addressing emerging, nuanced and
wide-reaching generative AI challenges.
Ensure conditions for knowledge-
sharing feedback loops
Knowledge-sharing requires nurturing of feedback
loop conditions and proactive examination of
challenges to those conditions that may prevent
stakeholders from meaningfully participating, as
described in Figure 2 and Table 6.
Feedback loop conditions for effective multistakeholder participation
F I G U R E 2
Feedback loop
conditions
Trustworthy
Communicative
Representative
Independent
Consistent
Transparent

Challenges impacting feedback loop conditions (non-exhaustive)
TA B L E 6
Stakeholder challenges Considerations for policy-makers
Trustworthy Industry may be wary of sharing models openly for fear
of divulging trade secrets or exposure to legal liabilities.
– Policy-makers should provide safe harbour
provisions and ensure discretion.
– To ensure mutual benefit, all participants should
be willing to share insights while preventing
privileged access.
Communicative CSOs (that are more fluent in social impacts), industry
(more fluent in technology) and government (more
fluent in policy) may have difficulty understanding each
other. Further complicating the issue, CSOs may often
examine topics through the lens of human rights,
whereas industry does so through risks.
– Policy-makers should use professional facilitators,
invest in structured support for participation
across sociotechnical conversations and increase
incorporation of rights protections in frameworks
(including in risk-based frameworks).
Representative Broad participation of actors is needed but can
be difficult to coordinate, and its inputs can be
hard to synthesize.
– Policy-makers could layer broad input models (e.g.
written input) over narrow models (e.g. roundtable).
– They could set ample time for input review
and synthesis.
Independent The public may be concerned about regulatory capture
or undue influence in boards or research partnerships.
– Policy-makers could set term limits for participation
in boards.
– They could make disclosure of extent of
industry participation in research collaborations
a requirement.
Consistent Sporadic touchpoints can leave non-industry
participants playing catch-up on technological
advances, and cause non-government
participants to lag behind on policy changes.
– Policy-makers should align on frequency
expectations and coordinate multiple
feedback loops.
Transparent Participants and the public may be concerned that
some stakeholders yield greater influence.
– Policy-makers could include equitable sectoral
representation and provide transparency on
feedback review processes with strengthened
whistleblower protections.
Governments will need to coordinate multiple
feedback models simultaneously to build holistic
knowledge-sharing across issues and timelines
(e.g. timing of AI model releases and legislative
calendars), and to account for long-standing and
emerging issues. Layering models is also necessary
to address limited resources. For example, calls
for inputs, which enable insights from numerous
stakeholders, can require substantial resources
to meaningfully review. Governments may consider
combining routine calls for input with more narrow
feedback mechanisms, such as advisory boards.
The boards themselves may conduct interviews
and roundtables to broaden representation of the
insights they share with policy-makers.
In designing feedback loops, policy-makers should
also consider how non-government stakeholders
have limited resources. It is also crucial to explore
how to simplify participation by, for instance,
reducing unnecessary complexities in calls-for-
input forms or merging similar calls for input from
different agencies to reduce time requirements
from participants.
Encourage interdisciplinary
innovation
Generative AI innovation is built upon
interdisciplinary research. For example, the
development of ImageNet, a database that proved
the importance of big data in training, emerged
from the cross-pollination of ideas from linguistics,
psychology, computer science and adjacent
fields.55
Despite the importance of interdisciplinary
collaboration to generative AI innovation and
addressing generative AI’s sociotechnical
challenges, industry and academia do not
sufficiently cultivate environments that support this
approach. Within private-sector tech companies,
social scientists and humanities experts are often
a fraction of the team. Despite maintained multi-
disciplinary faculties within academic institutions,
there are strong incentives for researchers
to publish within discipline-specific journals,
consequently encouraging isolated research. Policy-
makers should consider levers to address these
challenges, such as targeted academic research
grants with interdisciplinary requirements or financial
subsidies for interdisciplinary industry RD.

Lead by example with
responsible AI in public initiatives
Making use of AI, including generative AI, may
improve governments’ productivity, responsiveness
and accountability.56
However, its adoption requires
responsible design, development, deployment
and use, given its impact on individuals and society.
Setting an example of responsible AI practices in
government (including responsible procurement
and acquisitions) could help to establish responsible
AI norms57
and secure the participation of industry,
academia and civil society in creating a robust,
responsible AI network. The City Algorithm
Register, adopted across several cities in Europe,
enables citizens to review algorithms employed by
government agencies in public services, enhancing
public oversight.58
Jurisdictions such as Australia59
and the US60
have published internal policies for
government AI practices aimed at advancing
responsible innovation and managing risks.

Plan future
Generative AI governance demands
preparedness, agility and international
cooperation to address evolving sociotechnical
impacts and global challenges.
Generative AI’s capabilities are rapidly evolving
alongside other technologies and interacting
with changing market forces, user behaviour and
geopolitical dynamics. Bringing ongoing clarity
to generative AI’s changing short- and long-term
uncertainties is critical for effective governance.
Pillar 3
Government challenges and actions to keep pace with generative AI
TA B L E 7
Compounding challenges Strategic actions
Limited resources and expertise: Governments may struggle
to prioritize investment in building state-of-the-art AI and generative
AI expertise compared to other pressing needs.
Targeted investments and upskilling: Governments should
be deliberate with limited resources in upskilling and hiring.
Rapid evolution: Governments may lack sufficient proximity to, and
awareness of, generative AI evolution and adoption to effectively
approximate sociotechnical impacts.
Horizon scanning: Governments should monitor emerging
and converging generative AI capabilities and evolving
interactions with society.
Uncertain futures: Technology, society and geopolitical
uncertainties are outpacing traditional upskilling practices
and policy development cycles.
Strategic foresight: Governments should ensure resilience
though exercises that inform anticipatory policy.
Slow mechanisms: Government decision-making can
be slow by design (e.g. due to separation of powers and
oversight) or complicated by administrative procedures.
Impact assessments and agile regulations: Governments should
prepare for the downstream effects of regulation and introduce agile
dynamics into decision-making processes.
Global fragmentation: Limited resource-sharing and segregated
jurisdictional governance activity can paralyse domestic investment
and policy, and create non-interoperable international markets.
International cooperation: Governments should drive collective
action to keep pace with generative AI innovation through
harmonized standards and risk definitions, and sharing
of knowledge and infrastructure.
3.1
Targeted investments and upskilling
– Training on use: Ensure officials who use
generative AI technologies are trained in their
varied capabilities and limitations.
– Training on procurement: Ensure officials who
work with vendors are equipped to assess and
test the AI capabilities of a product.
– Adaptive upskilling: Collaborate with industry
and academia on adaptive upskilling of
government in AI and foundational digital literacy.
– Strategic hiring: Recruit specialists for
positions identified with amplified impact and,
with limited resources, consider prioritizing
sectors and use cases, for example, based on
risk or domestic economic factors.
– Hiring vs upskilling: Consider how to
appropriately balance hiring AI experts
with AI upskilling of sector-specific experts
(e.g. in agriculture and health).
– AI body: Carefully consider the need and
scope of an AI-specific body or authority (see
“Expansion of existing regulatory authority
competencies” under section 1.4).
– Guidance: Examine where frameworks can
be applied across sectors and where investment
is needed for sector-specific guidance.

3.2
Horizon scanning
To anticipate and navigate novel risks and
challenges posed by frontier generative AI,
governance frameworks must continuously examine
the horizon of generative AI innovation, including:
– Emergence of new generative AI capabilities
– Convergence of generative AI with
other technologies
– Interactions with generative AI technologies
Documented, planned or forecasted emergence,
convergence and interaction patterns can yield
new waves of economic opportunities and novel
approaches to addressing social and environmental
challenges. Ongoing monitoring of opportunities
and risks is critical to steering generative AI
towards being a technology that benefits society.
Multistakeholder knowledge-sharing (see Table 8)
can enable informed horizon scanning.
Policy-makers should collaborate with industry to
provide guidance on where disclosure of identified
risks is needed and support oversight mechanisms
to ensure compliance.
Emergence
As developers scale up generative AI models, the
latter may exhibit qualitative changes in capabilities
that do not present in smaller models. Such
unexpected capabilities may include potentially
risk-inducing abilities such as adaptive persuasion
strategies, “power-seeking behaviours” to accrue
resources and authority, and autonomous replication,
adaptation and long-term planning capabilities.
These emergent model properties must inspire
appropriate governance benchmarks to effectively
address unpredictable powers and potential pitfalls.
Generative AI emergent capabilities (non-exhaustive)
TA B L E 8
Category Example use Example risks
Considerations for
policy-makers
Multimodal
generative AI
Systems that
synthesize and
generate outputs
across diverse
data types and
sensory inputs
Data analysed from radars,
cameras, light detection and
ranging (LiDAR), sensors and
global positioning systems
(GPS) in a safety-critical system
(like a self-driving vehicle)
to predict the behaviour of
surrounding vehicles and
pedestrians more accurately61
– Compounded data
manipulation across input types
– Amplification of potential flaws,
biases and vulnerabilities
– Novel systemic failures
– Exacerbated societal disparities
– Scaled and difficult-to-detect
mis/disinformation
– Novel persuasion techniques
– Focus on data integrity and
secure-by-design frameworks,
model architecture disclosures,
responsible system design
and impact assessment
in public sectors
– Examine readiness of existing
policies and, if necessary, amend
to address emerging privacy,
security, safety, fairness, and
IP rights and accountability
Multi-agent
generative AI
AI systems involving
multiple agents that
autonomously pursue
complex goals with
minimal supervision
Swarms of drones deployed for
military and security purposes62
– Increased unpredictability
and control complexity
– Added accountability
complexity
– Challenges to traditional
scenario planning and
risk management
– Potential for cascading failures
– Novel adversarial attacks
– Develop guidelines for design and
testing focused on robustness,
security, safety, transparency,
traceability and explainability
– Establish accountability
frameworks
Embodied
generative AI
AI systems embodied
within physical entities
such as robotics and
devices capable of
interacting with the
real world
General-purpose humanoid
robot with neural network-
powered manual dexterity
and ChatGPT 4’s visual
and language intelligence63
– Physical safety risks from
control system failures
– Security issues from malicious
use of such systems
– Novel physical manifestations
of hallucinations
– Implement safety standards
and security benchmarks
– Encourage voluntary industry
reviews and supplement with
certification and audit practices,
where appropriate

Convergence
As a powerful general-purpose technology,
generative AI can amplify other technologies,
old and new, exposing complex governance
challenges. For example, social media is under
scrutiny due to its potential to distribute harmful
AI-generated deepfakes,64
such as non-consensual
pornography65
– including CSAM66
– and election
disinformation.67
Looking ahead, the convergence
of generative AI with advanced technologies can
pose unprecedented opportunities and risks,
as both the technologies and their governance
frameworks are in the early stages.
Generative AI convergence with advanced technologies (non-exhaustive)
TA B L E 9
Category Example uses Example risks
Considerations for
policy-makers
Synthetic biology
Generative AI is increasingly used
in developing artificial analogues of
natural processes, e.g. generation
of genome sequences and
cellular images, and simulations
of genes and proteins. It is also
used in building “virtual labs”68
that can mitigate space and
hazardous waste of real-world
experimentations.
– Unintended ecological
consequences
– Gain-of-function research
giving naturally occurring
diseases new symptoms or
capabilities like resiliency to
medical treatments
– Biosecurity risks and
biological warfare
– Novel ethical implications
– Robust bioethical frameworks
– Tracking of the building and
operation of various high-
security disease labs globally
– Restrictions on
high-risk research
– Strict containment protocols
– International collaboration on
safety standards
– Refocusing of existing biological
control laws
Neurotechnology
Progress in generative AI,
neuroscience and the development
of brain-computer interfaces offers
potential for increasing scientific
discoveries, enabling paralysed
individuals with communication, as
well as addressing the burden of
neurological disease and mental
illnesses such as attention deficit
hyperactivity disorder (ADHD), post-
traumatic stress disorder (PTSD)
and severe depression.
– Intentional abuse
– Use in lethal autonomous
weapon systems
– Cognitive enhancement by
brain-computer interfaces can
amplify existing inequities
– Behaviour modification and
manipulation
– Enfeeblement
– Review of privacy approaches
that consider cognitive freedom,
liberty and autonomy, and the
establishment of new digital
rights, if necessary
– Establishment of assessment
standards for model or
neuroscientific accounts
of disease on individuals,
communities and society
– Internationally harmonized
ethical standards for biological
material and data collection
– Examination of moral
significance of neural
systems under development
in neuroscience research
laboratories
– Context specification for
neuroscientific technology use
and deployment
Quantum
computing
Through optimizing code,
generative AI may improve the
design of hardware and quantum
computing circuits, which are
intended to solve problems too
complex for classical computing.
Quantum computing may
accelerate generative AI training
and inference and optimize
parameter exploration.
– Advanced models beyond
human comprehension
– Impact on the environment
due to increased energy and
resource demands
– Review of legal provisions
for controlled innovation that
balance pace and safety
without hindering progress
– Incentivization of sustainable
practices and energy-efficient
technologies
– Consideration of measures
such as investing in research
to strengthen the security and
privacy of these systems

Emotional entanglement
Emotional AI aims to recognize, interpret and
respond to human emotions, potentially improving
human-computer interactions. As generative
AI applications become more complex and
computationally powerful, the risk of emotional
reliance between humans and generative AI
applications tends to increase.69
Risks include
dependency, privacy issues, coercion or
manipulation leading to safety or psychological
risks.70
Such issues are exemplified by cases of
users claiming that AI companies are interfering
with their romantic relationships with chatbots.71
The gravity of these phenomena is already evident
in society, as seen in the case of a man who
reportedly “ended his life following a six-week-
long conversation about the climate crisis with an
AI chatbot”.72
Careful consideration of the ethical
implications by policy-makers and legislators to
ensure responsible AI use will be necessary.73
Synthetic data feedback loops
Human-created content scraped from the internet
has been crucial in the training of large-scale
machine learning, but this reliance is at risk due
to the increasing prevalence of synthetic data
generated by AI models.74
Training models with
synthetic data could lead to “model collapse”,
where the quality of the generated content
degrades over successive iterations, causing
the performance of the models to deteriorate.75
Policy-makers, in collaboration with industry,
academia and CSOs, will need to consider how
to stabilize these systems with human feedback,
preserve human-created knowledge systems and
incentivize the production and curation of high-
quality data. Such considerations will need to be
balanced against the requirements of substantial
storage and processing resources, potentially
impacting policy efforts related to sustainability.
As generative
AI applications
become more
complex and
computationally
powerful, the risk
of emotional
reliance between
humans and
generative AI
applications tends
to increase.
Interactions
Today, the integration of generative AI technologies
into personal AI virtual assistants and companions
raises new challenges that emerge from human
interaction with and emotional reliance on these
technologies. This issue highlights the need
for responsible implementation, privacy, data
protection and ethical human-AI interaction.
For example, rapid advances and interactions
with generative AI-enabled neurotechnology could
become mainstream for many children, largely
as consumer electronic devices that are not
subject to rigorous oversight in clinical settings.
The advancement and proliferation of voice
chatbots, often with female-presenting voices,
raise concerns about reinforced gender biases
and stereotyping. Responsible and ethical
development and regulation of these technologies,
grounded in human rights, must therefore
be an area of attention across stakeholder groups.

Agile and flexible regulation is essential in AI to
address evolving financial, economic and social
impacts. Policy-makers must consider diverse
stakeholder input to account for varied sectoral
and community short- and long-term impacts.
Governments should also study varied agile
practices emerging globally and assess jurisdictional
fit. For example, they should consider regulatory
sandboxes for testing prior to broad deployment.
Another approach is “complex adaptive regulations”,
which are designed to respond to the effects they
create and require defined goals, success metrics
and thresholds for how regulations will adapt to their
own impacts.
Governmental structures can adopt the dynamics
of tech companies to become more agile through:
1) a risk-based approach, 2) regular review of
technology and marketplace challenges, 3) agile
response to challenges,78
and 4) review of response
effects and adaptation.79
Still, agile governance
should not come at the expense of oversight or
separation of powers, nor without regard to human
rights and rights-based frameworks that ensure
that generative AI development and deployment
align with societal values and norms. Governments
should avoid adopting a “move fast and break
things” form of hyper-agility that has been criticized
for prioritizing go-to-market testing over mitigation
of harmful consequences.
3.4
Impact assessments and agile regulations
Often, individuals and institutions rely on a default
set of assumptions about the future. However, the
future is inherently uncertain. For a technology as
rapidly evolving (and with such complex geopolitics)
as generative AI, unexamined assumptions can lead
to miscalculations in governance.
Strategic foresight is a set of methodologies and
tools that allow for an organized, scientific approach
to thinking about, and preparing for, the future.
Adoption of strategic foresight helps governments
be agile – to move beyond assumptions of the
future, systematically explore critical uncertainties,
envisage potential solutions and risks, sandbox
new ideas and articulate alternate visions of
successful futures.
Strategic foresight has been adopted successfully
by various governments. For example, in Finland,
the Government Report on the Future sets
parameters for long-term planning and decision-
making.76
In the United Arab Emirates, the Dubai
Future Foundation (DFF) leads 13 councils,77
each of
which convenes government directors and experts
to investigate the future of different sectors or issue
areas (such as AI), and to identify the governance
and capacity needed to drive positive change.
Although strategic foresight initiatives vary, best
practices include:
– Guided: Use models or prompts to guide
exercises, e.g. use scenario planning matrices
to consider potential futures across axes of
critical uncertainties.
– Consistent: Plan exercises on a recurring basis
and identify organizational champions.
– Multistakeholder: Engage cross-functional
internal and external stakeholders to mitigate
biases and map multiple possible futures.
– Transparent: Track and measure adoption –
for example, in Dubai, a numerical scale was
developed to rank the effectiveness of each
agency in integrating strategic foresight and
rankings were then shared to increase healthy
competition and incentivize adoption.
3.3
Strategic foresight
25

Key areas requiring international cooperation between jurisdictions
TA B L E 1 0
Standards
– Standards can help make abstract AI principles actionable, are more agile than regulations and can bolster
global resilience while regulation processes are underway.
– They are critical to regulatory interoperability.81
– Quality assurance techniques and technical standards support cross-border trade. Provisions in free trade
agreements (FTAs) are needed to address challenges facing AI innovators. Testing certifications should
be interoperable where possible.
– Anticipatory standards require increased inclusion of CSOs and academia, and coordination
of standards bodies.82
Safety
– Strengthened RD of safety techniques and evaluation tools is key to resilience.
– It is crucial to coordinate AI safety institutes to maximize limited resources. An agreement signed
by various jurisdictions at the AI Seoul Summit on a network of institutes is promising.83
– It is additionally necessary to ensure that long-term risks are not prioritized at the expense of identified
present AI harms.84
Risks
– Establishing mutual understanding of 1) taxonomy of risks, 2) definition and scope of mitigating risks,
and 3) approaches is necessary to evaluate, quantify and determine if a model/application meets
the risk mitigation threshold.
– It is essential to embrace jurisdictional variability on risk tolerance and ethical principles,85
while advancing risk
management interoperability. This can be achieved by considering how standards may apply across high-risk
cases while leaving the definition of “high-risk” to jurisdictions.
– Collaboration across sectors is crucial for proactively identifying generative AI opportunities and risks (including
critical-, systemic- and infrastructure-related). This could be achieved via a dedicated international observatory.
Prohibitions
– Lack of alignment on prohibitions increases the likelihood of generative AI misuse by state or non-state actors
with severe global consequences.
– Collaboration on treaties or other norm-building mechanisms is needed to establish clear prohibitions on specific
forms of generative AI research, development, deployment and use.
Knowledge-
sharing
– Participation in a platform, such as a global governance sandbox, enables the sharing of best practices, case
studies (e.g. technical, ethical and legal) and tools that allow stakeholders to implement informed governance.
Infrastructure
– Many jurisdictions have limited access to compute and high-quality data for training and fine-tuning, leading
to reliance on models prone to error in local languages or contexts. Even open models are not easily fine-tuned
to a new language due to underlying tokenization.
– Examination of opportunities for multilateral sharing, or shared ownership, of compute and data, alongside the
mitigation of bad-actor access or certain other uses, e.g. military.
– Developed countries should prioritize sharing resources, expertise and best practices to enable global majority
countries to build their AI capabilities and participate effectively in international forums.
The current international discussions on generative AI
governance frequently lack meaningful participation
from global majority countries. This can create
significant knowledge gaps about the risks,
opportunities and prospects of the generative AI
supply chain in those underrepresented regions.80
Principles and frameworks developed without
their input may prove ineffective or even harmful.
Unaddressed, these tensions could lead to a
fragmentation of the global generative AI community
into segregated, non-interoperable spheres.
Thus, international cooperation is essential in six
areas (see Table 10) to harness the benefits of
generative AI while managing its dangers equitably.
This can be achieved through bilateral, regional and
broader international mechanisms of cooperation,
like those advanced by the World Economic Forum,
the United Nations (UN), Group of 20 (G20), the
Organisation for Economic Co-operation and
Development (OECD) and the African Union High
Level Panel on Emerging Technologies (APET).
3.5
International cooperation

Conclusion
This paper is intended to provide policy-makers
and regulators with a detailed, practical and
implementable generative AI governance
framework. Generative AI, like other technologies,
is not neutral – it touches upon shared values
and fundamental rights. Before introducing new
AI regulations, it is crucial to evaluate the current
regulatory landscape and enhance coordination
among sectoral regulators to mitigate generative
AI-induced tensions. Existing regulatory authorities
should be assessed for their capability to respond
to emerging generative AI challenges, and the
trade-offs of a distributed governance approach
versus a single dedicated agency should be
considered. A comprehensive whole-of-society
governance strategy should address industry,
civil society and academic challenges, promoting
cross-sector collaboration and interdisciplinary
solutions. Looking ahead, future strategies need
to account for resource limitations and global
uncertainties, with adaptable foresight mechanisms
and international cooperation through standardized
practices and shared knowledge. By adopting a
harmonized approach, generative AI challenges can
be addressed more effectively at a global level.

Contributors
Lead authors
Rafi Lazerson
Responsible AI Specialist, Accenture; Project
Fellow, AI Governance Alliance
Manal Siddiqui
Responsible AI Manager, Accenture; Project Fellow,
AI Governance Alliance
Karla Yee Amezaga
Lead, Data Policy and AI, World Economic Forum
Samira Gazzane
Policy Lead, Artificial Intelligence
and Machine Learning
Accenture
Patrick Connolly
Responsible AI Research Manager
Kathryn White Krumpholz
Managing Director, Innovation Incubation;
Executive Fellow, AI Governance Alliance
Andrew J.P. Levy
Chief Corporate and Government Affairs Officer
Valerie Morignat
Responsible AI Senior Manager, Accenture;
Project Fellow, AI Governance Alliance
Charlie Moskowitz
Government Relations Senior Manager
Ali Shah
Managing Director, Responsible AI; Executive
Fellow, AI Governance Alliance
Dikshita Venkatesh
Responsible AI Research Senior Analyst;
Project Fellow, AI Governance Alliance
This paper is a combined effort based on numerous interviews, discussions, workshops and
research. The opinions expressed herein do not necessarily reflect the views of the individuals
or organizations involved in the project or listed below. Sincere thanks are extended to those who
contributed their insights via interviews and workshops, as well as those not captured below.
Acknowledgements
Sincere appreciation is extended to the following
working group members, who spent numerous
hours providing critical input and feedback on
the drafts. Their diverse insights are fundamental
to the success of this work.
Lovisa Afzelius
Chief Executive Officer, Apriori Bio
Hassan Al-Darbesti
Adviser to the Minister and Director, International
Cooperation Department, Ministry of Information
and Communication Technology (ICT) of Qatar
Uthman Ali
Global Responsible AI Officer, BP
Jason Anderson
General Counsel, Vice-President and Corporate
Secretary, DataStax
Norberto Andrade
Professor and Academic Director, IE University
Jesse Barba
Head, Government Affairs and Policy, Chegg
Richard Benjamins
Co-Founder and Chief Executive Officer, OdiseIA
Saqr Binghalib
Executive Director, Artificial Intelligence, Digital
Economy and Remote Work Applications Office of
the United Arab Emirates
Anu Bradford
Professor, Law, Columbia Law School
Daniela Braga
Founder and Chief Executive Officer, Defined.ai
Michal Brand-Gold
Vice-President General Counsel, ActiveFence

Adrian Brown
Executive Director, Center for Public Impact
Melika Carroll
Head, Global Government Affairs and Public
Policy, Cohere
Winter Casey
Senior Director, SAP
Daniel Castano Parra
Professor, Law, Universidad Externado de Colombia
Neha Chawla
Senior Corporate Counsel, Infosys
Simon Chesterman
Senior Director, AI Governance, AI Singapore,
National University of Singapore
Quintin Chou-Lambert
Office of the UN Tech Envoy, United Nations
Melinda Claybaugh
Director of Privacy Policy, Meta Platforms
Frincy Clement
Head, North America Region, Women in AI
Magda Cocco
Head, Practice Partner, Information, Communication
and Technology, Vieira de Almeida and Associados
Amanda Craig
Senior Director, Responsible AI Public Policy,
Microsoft
Renée Cummings
Data Science Professor and Data Activist in
Residence, University of Virginia
Gerard de Graaf
Senior EU Envoy for Digital to the US, European
Commission
Nicholas Dirks
President and Chief Executive Officer, The New York
Academy of Sciences
Mark Esposito
Faculty Affiliate, Harvard Center for International
Development, Harvard Kennedy School and
Institute for Quantitative Social Sciences
Nita Farahany
Robinson O. Everett Professor of Law and
Philosophy, Duke University; Director, Duke Science
and Society
Max Fenkell
Vice-President, Government Relations, Scale AI
Kay Firth-Butterfield
Chief Executive Officer, Good Tech Advisory
Katharina Frey
Deputy Head, Digitalisation Division, Federal
Department of Foreign Affairs (FDFA) of Switzerland
Alice Friend
Head, Artificial Intelligence and Emerging Tech
Policy, Google
Tony Gaffney
President and Chief Executive Officer,
Vector Institute
Eugenio Garcia
Director, Department of Science, Technology,
Innovation and Intellectual Property (DCT),
Brazilian Ministry of Foreign Affairs (Itamaraty)
Urs Gasser
Dean, TUM School of Social Sciences and
Technology, Technical University of Munich
Justine Gauthier
Director, Corporate and Legal Affairs, MILA -
Quebec Artificial Intelligence Institute
Debjani Ghosh
President, National Association of Software
and Services Companies (NASSCOM)
Danielle Gilliam-Moore
Director, Global Public Policy, Salesforce
Anthony Giuliani
Global Head of Operations, Twelve Labs
Brian Patrick Green
Director, Technology Ethics, Markkula Center
for Applied Ethics, Santa Clara University
Samuel Gregory
Executive Director, WITNESS
Koiti Hasida
Director, Artificial Intelligence in Society Research
Group, RIKEN Center for Advanced Intelligence
Project, RIKEN
Dan Hendrycks
Executive Director, Center for AI Safety
Benjamin Hughes
Senior Vice-President, Artificial Intelligence (AI)
and Real World Data (RWD), IQVIA
Marek Jansen
Senior Director, Strategic Partnerships and Policy
Management, Volkswagen
Jeff Jianfeng Cao
Senior Research Fellow, Tencent Research Institute
Sam Kaplan
Assistant General Counsel and Senior Director,
Palo Alto Networks

Kathryn King
General Manager, Technology and Strategy, Office
of the eSafety Commissioner Australia
Edward S. Knight
Executive Vice-Chairman, Nasdaq
James Laufman
Executive Vice-President, General Counsel and
Chief Legal Officer, Automation Anywhere
Alexis Liu
Head, Legal, Weights and Biases
Caroline Louveaux
Chief Privacy and Data Responsibility Officer,
Mastercard
Shawn Maher
Global Vice-Chair, Public Policy, EY
Gevorg Mantashyan
First Deputy Minister, High-Tech Industry, Ministry of
High-Tech Industry of Armenia
Gary Marcus
Chief Executive Officer, Center for Advancement of
Trustworthy AI
Gregg Melinson
Senior Vice-President, Corporate Affairs, Hewlett
Packard Enterprise
Robert Middlehurst
Senior Vice-President, Regulatory Affairs,
e International
Satwik Mishra
Executive Director, Centre for Trustworthy
Technology, Centre for the Fourth
Industrial Revolution
Casey Mock
Chief Policy and Public Affairs Officer, Center for
Humane Technology
Chandler Morse
Vice-President, Corporate Affairs, Workday
Henry Murry
Vice-President, Government Relations, C3 AI
Miho Naganuma
Senior Executive Professional, Digital Trust Business
Strategy Department, NEC
Didier Navez
Senior Vice-President, Data Policy Governance,
Dawex
Dan Nechita
Former Head of Cabinet, MEP Dragoș Tudorache,
European Parliament (2019-2024)
Jessica Newman
Director, AI Security Initiative, Centre for Long-Term
Cybersecurity, UC Berkeley
Michael Nunes
Vice-President, Payments Policy, Visa
Bo Viktor Nylund
Director, UNICEF Innocenti Global Office of
Research and Foresight, United Nations Children’s
Fund (UNICEF)
Madan Oberoi
Executive Director, Technology and Innovation,
International Criminal Police Organization
(INTERPOL)
Florian Ostmann
Head, AI Governance and Regulatory Innovation,
The Alan Turing Institute
Marc-Etienne Ouimette
Lead, Global AI Policy, Amazon Web Services
Timothy Persons
Principal, Digital Assurance and Transparency of US
Trust Solutions, PwC
Tiffany Pham
Founder and Chief Executive Officer, Mogul
Oreste Pollicino
Professor, Constitutional Law, Bocconi University
Catherine Quinlan
MA Legal Integration Executive, IBM
Roxana Radu
Associate Professor of Digital Technologies
and Public Policy, Blavatnik School of
Government; Hugh Price Fellow, Jesus College
University of Oxford
Martin Rauchbauer
Co-Director and Founder, Tech Diplomacy Network
Alexandra Reeve Givens
Chief Executive Officer, Center for Democracy and
Technology
Philip Reiner
Chief Executive Officer, Institute for Security and
Technology
Andrea Renda
Senior Research Fellow, Centre for European Policy
Studies (CEPS)
Rowan Reynolds
General Counsel and Head of Policy, Writer
Sam Rizzo
Head, Global Policy Development, Zoom Video
Communications

John Roese
Global Chief Technology Officer, Dell Technologies
Nilmini Rubin
Chief Policy Officer, Hedera Hashgraph
Arianna Rufini
ICT Adviser to the Minister, Ministry of Enterprises
and Made in Italy
Crystal Rugege
Managing Director, Centre for the Fourth Industrial
Revolution Rwanda
Joaquina Salado
Head, AI Ethics, Telefónica
Idoia Salazar
Professor, CEU San Pablo University
Nayat Sanchez-Pi
Chief Executive Officer, INRIA Chile
Mark Schaan
Deputy Secretary to the Cabinet (Artificial
Intelligence), Privy Council Office, Canada
Thomas Schneider
Ambassador and Director of International Affairs,
Swiss Federal Office of Communications, Federal
Department of the Environment, Transport, Energy
and Communications (DETEC)
Robyn Scott
Co-Founder and Chief Executive Officer, Apolitical
Var Shankar
Affiliate, Governance and Responsible AI Lab
(GRAIL Lab), Purdue University
Navrina Singh
Founder and Chief Executive Officer, Credo AI
Scott Starbird
Chief Public Affairs Officer, Databricks
Uyi Stewart
Chief Data and Technology Officer, data.org
Charlotte Stix
Head, AI Governance, Apollo Research
Arun Sundararajan
Harold Price Professor, Entrepreneurship and
Technology, Stern School of Business, New York
University
Nabiha Syed
Executive Director, Mozilla Foundation
Patricia Thaine
Co-Founder and Chief Executive Officer, Private AI
V Valluvan Veloo
Director, Manufacturing Industry, Science and
Technology Division, Ministry of Economy, Malaysia
Ott Velsberg
Government Chief Data Officer, Ministry of
Economic Affairs and Information Technology
of Estonia
Miriam Vogel
President and Chief Executive Officer, Equal AI
Takuya Watanabe
Director, Software and Information Service Industry
Strategy Office, Ministry of Economy, Trade and
Industry Japan
Andrew Wells
Chief Data and AI Officer, NTT DATA
Denise Wong
Assistant Chief Executive, Data Innovation and
Protection Group, Infocomm Media Development
Authority of Singapore
Kai Zenner
Head, Office and Digital Policy Adviser, MEP Axel
Voss, European Parliament
Arif Zeynalov
Transformation Chief Information Officer, Ministry of
Economy of the Republic of Azerbaijan
Sincere appreciation is also extended to the
following individuals who contributed their insights
for this report.
Basma AlBuhairan
Managing Director, Centre for the Fourth Industrial
Revolution, Saudi Arabia
Abdulaziz AlJaziri
Deputy Chief Executive Officer and Chief Operations
Officer, Dubai Future Foundation
Dena Almansoori
Group Chief AI and Data Officer, e
Daniela Battisti
Senior Advisor and International Relations Expert,
Department for Digital Transformation, Italian
Presidency of the Council of Ministers
Daniel Child
Manager, Industry Affairs and Engagement,
Office of the eSafety Commissioner Australia
Valeria Falce
Full Professor of Economic Law, Senior Advisor and
Legal Expert, Department for Digital Transformation,
Italian Presidency of the Council of Ministers

Lyn Jeffery
Distinguished Fellow and Director, Institute for the
Future (IFTF)
Japan External
Trade Organization
Genta Ando
Executive Director and Project Fellow,
Hitachi America
Daisuke Fukui
Senior Researcher and Project Fellow,
Minos Bantourakis
Head, Media, Entertainment and Sport Industry
Maria Basso
Portfolio Manager, Digital Technologies
Agustina Callegari
Lead, Global Coalition for Digital Safety
Daniel Dobrygowski
Head, Governance and Trust
Karyn Gorman
Communications Lead, Metaverse Initiative
Ginelle Greene-Dewasmes
Lead, AI and Energy
Bryonie Guthrie
Lead, Foresight and Organizational Transformation
Jill Hoang
Lead, AI and Digital Technologies
Devendra Jain
Lead, Artificial Intelligence, Quantum Technologies
Jenny Joung
Specialist, Artificial Intelligence
Connie Kuang
Lead, Generative AI and Metaverse Value Creation
Benjamin Larsen
Lead, Artificial Intelligence and Machine Learning
Na Na
Lead, Advanced Manufacturing
and Artificial Intelligence
Chiharu Nakayama
Lead, Data and Artificial Intelligence
Hannah Rosenfeld
Specialist, Artificial Intelligence
Nivedita Sen
Initiatives Lead, Institutional Governance
Stephanie Smittkamp
Coordinator, AI and Data
Stephanie Teeuwen
Specialist, Data and AI
Kenneth White
Manager, Communities and Initiatives,
Institutional Governance
Hesham Zafar
Lead, Business Engagement
Production
Louis Chaplin
Editor, Studio Miko
Laurence Denmark
Creative Director, Studio Miko
Cat Slaymaker
Designer, Studio Miko

Endnotes
1. Bielefeldt, H., Weiner, M. (2023). Declaration on the Rights of Persons Belonging to National or Ethnic, Religious and
Linguistic Minorities. United Nations. https://legal.un.org/avl/pdf/ha/ga_47-135/ga_47-135_e.pdf.
2. United Nations (UN). (1990). The United Nations Convention on the Rights of the Child.
https://www.unicef.org.uk/wp-content/uploads/2010/05/UNCRC_PRESS200910web.pdf.
3. United Nations Office on Drugs and Crime. (n.d.). Ad Hoc Committee to Elaborate a Comprehensive International
Convention on Countering the Use of Information and Communications Technologies for Criminal Purposes.
https://www.unodc.org/unodc/en/cybercrime/ad_hoc_committee/home.
4. United Nations. (1992). United Nations Framework Convention on Climate Change.
https://unfccc.int/files/essential_background/background_publications_htmlpdf/application/pdf/conveng.pdf;
United Nations. (2015). Paris Agreement. https://unfccc.int/sites/default/files/english_paris_agreement.pdf.
5. Leslie, D., Burr, C., Aitken, M., Cowls, J., Katell, M., Briggs, M. (2021). Artificial intelligence, human rights, democracy,
and the rule of law: A primer. SSRN. https://doi.org/10.2139/ssrn.3817999.
6. World Economic Forum. (2023). Data Equity: Foundational Concepts for Generative AI.
https://www3.weforum.org/docs/WEF_Data_Equity_Concepts_Generative_AI_2023.pdf.
7. World Economic Forum. (2020). A New Paradigm for Business of Data.
https://www3.weforum.org/docs/WEF_New_Paradigm_for_Business_of_Data_Report_2020.pdf.
8. Van Bekkum, M., Zuiderveen Borgesius, F. (2023). Using sensitive data to prevent discrimination by artificial intelligence:
Does the GDPR need a new exception? Computer Law Security Review, vol. 48.
https://doi.org/10.1016/j.clsr.2022.105770.
9. Reisner, A. (2023). Generative AI Might Finally Bend Copyright Past the Breaking Point. The Atlantic.
https://www.theatlantic.com/technology/archive/2024/02/generative-ai-lawsuits-copyright-fair-use/677595/.
10. UK Government. (2023). Pro-innovation Regulation of Technologies Review—Digital Technologies.
https://assets.publishing.service.gov.uk/media/64118f0f8fa8f555779ab001/Pro-innovation_Regulation_of_Technologies_
Review_-_Digital_Technologies_report.pdf.
11. House of Lords Communications and Digital Committee. (2024). Large language models and generative AI.
https://publications.parliament.uk/pa/ld5804/ldselect/ldcomm/54/54.pdf.
12. Shan, S., Ding, W., Passananti, J., Wu, S., Zheng, H., Zhao, B. Y. (2024). Nightshade: Prompt-Specific Poisoning
Attacks on Text-to-Image Generative Models. Department of Computer Science, University of Chicago.
https://people.cs.uchicago.edu/~ravenben/publications/pdf/nightshade-oakland24.pdf.
13. Grynbaum, M. M., Mac, R. (2023). The Times Sues OpenAI and Microsoft Over AI Use of Copyrighted Work. The New
York Times. https://www.nytimes.com/2023/12/27/business/media/new-york-times-open-ai-microsoft-lawsuit.html.
14. Hays, K. (2023). Andreessen Horowitz would like everyone to stop talking about AI’s copyright issues, please.
Business Insider. https://www.businessinsider.com/marc-andreessen-horowitz-ai-copyright-2023-11.
15. Hoppner, T., Ufues, S. (2024). On the Antitrust Implications of Embedding Generative AI in Core Platform Services.
CPI Antitrust Chronicles, vol. 1, no. 12. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4904876.
16. European Commission. (2024, 23 July). Joint Statement on Competition in Generative AI Foundation Models and AI
Products [Press release]. https://competition-policy.ec.europa.eu/about/news/joint-statement-competition-generative-ai-
foundation-models-and-ai-products-2024-07-23_en.
17. Macko, M. S. (2024). Applying Data Minimization to Consumer Requests. California Privacy Protection Agency
Enforcement Division. https://cppa.ca.gov/pdf/enfadvisory202401.pdf.
18. Office of the Privacy Commissioner of Canada. (2023). Principles for responsible, trustworthy and privacy-protective
generative AI technologies. https://www.priv.gc.ca/en/privacy-topics/technology/artificial-intelligence/gd_principles_ai/.
19. Private AI. (n.d.). Background on Pll. https://docs.private-ai.com/introduction/#background.
20. Ontario Securities Commission. (2024). Data privacy and the Administrative Arrangement. https://www.osc.ca/en/about-
us/domestic-and-international-engagement/international-engagement/data-privacy-and-administrative-arrangement.
21. E-Safety Commissioner, Australian Government. (n.d.). Tech Trends Position Statement – Generative AI. Australian
Government. https://www.esafety.gov.au/industry/tech-trends-and-challenges/generative-ai.
22. Government of Canada. (2023, 12 October). Government of Canada launches consultation on the implications of
generative artificial intelligence for copyright [Press release]. https://www.canada.ca/en/innovation-science-economic-
development/news/2023/10/government-of-canada-launches-consultation-on-the-implications-of-generative-artificial-
intelligence-for-copyright.html.
23. US Copyright Office, Library of Congress. (2023). Copyright Registration Guidance: Works Containing Material Generated
by Artificial Intelligence 37 CFR Part 202. https://public-inspection.federalregister.gov/2023-05321.pdf.

24. Government of the United Kingdom. (2024). CMA seeks views on Microsoft’s partnership with OpenAI.
https://www.gov.uk/government/news/cma-seeks-views-on-microsofts-partnership-with-openai.
25. Atleson, M. (2023). Chatbots, deepfakes, and voice clones: AI deception for sale. Federal Trade Commission.
https://www.ftc.gov/business-guidance/blog/2023/03/chatbots-deepfakes-voice-clones-ai-deception-sale.
26. Competition and Market Authority. (2023). AI Foundation Models Review: Short Version.
https://assets.publishing.service.gov.uk/media/65045590dec5be000dc35f77/Short_Report_PDFA.pdf.
27. World Economic Forum. (n.d.). Digital Trust Framework. https://initiatives.weforum.org/digital-trust/framework.
28. Groves, L., Metcalf, J., Vecchione, B., Strait, A. (2024). Auditing Work: Exploring the New York City algorithmic bias
audit regime. ACM Digital Library. https://dl.acm.org/doi/10.1145/3630106.3658959.
29. Smith, B. (2023). How do we best govern AI? Microsoft on the Issues.
https://blogs.microsoft.com/on-the-issues/2023/05/25/how-do-we-best-govern-ai/.
30. Schrepel, T., Pentland, A. S. (2023). Competition Between AI Foundation Models: Dynamics and Policy
Recommendations. Massachusetts Institute of Technology Connection Science.
https://ide.mit.edu/wp-content/uploads/2024/01/SSRN-id4493900.pdf?x41178.
31. European Commission. (2024). Commission Decision Establishing the European AI Office.
https://digital-strategy.ec.europa.eu/en/library/commission-decision-establishing-european-ai-office.
32. World Economic Forum. (2024). AI Governance Alliance: Briefing Paper Series.
https://www.weforum.org/publications/ai-governance-alliance-briefing-paper-series/.
33. National Institute of Standards and Technology (NIST). (2024). AI Risk Management Framework.
https://www.nist.gov/itl/ai-risk-management-framework.
34. Marcus, G. (n.d.). AI Took My Career! [Broadcast].
https://podcasts.apple.com/gb/podcast/ai-took-my-career/id1532110146?i=1000624493662.
35. World Economic Forum. (2024). Responsible AI Playbook for Investors.
https://www3.weforum.org/docs/WEF_Responsible_AI_Playbook_for_Investors_2024.pdf.
36. The Forum’s AI Governance Alliance is currently researching energy resources as part of the work of the AI Transformation
of Industries pillar of work. Publications on this important topic will be released in coming months
37. Personal Data Protection Commission, Singapore. (n.d.). Advisory Guidelines on use of Personal Data in AI
Recommendation and Decision Systems. https://www.pdpc.gov.sg/guidelines-and-consultation/2024/02/advisory-
guidelines-on-use-of-personal-data-in-ai-recommendation-and-decision-systems.
38. Maslej, N. et al. (2024). The AI Index 2024 Annual Report. Institute for Human-Centered AI, Stanford University.
https://aiindex.stanford.edu/wp-content/uploads/2024/05/HAI_AI-Index-Report-2024.pdf.
39. National Institute of Standards and Technology (NIST). (n.d.). Generative AI: Text-to-Text (T2T).
https://ai-challenges.nist.gov/t2t.
40. International Standards Organization. (2023). ISO/IEC 42001:2023. https://www.iso.org/standard/81230.html.
41. Li, F.-F. (2023). Governing AI Through Acquisition and Procurement. Stanford Institute for Human-Centered Artificial
Intelligence (HAI), Stanford University. https://hai.stanford.edu/sites/default/files/2023-09/Fei-Fei-Li-Senate-Testimony.pdf.
42. European Commission. (2024). Living guidelines on the responsible use of generative AI in research.
https://research-and-innovation.ec.europa.eu/document/2b6cf7e5-36ac-41cb-aab5-0d32050143dc_en.
43. World Economic Forum. (2024). Shaping the Future of Learning: The Role of AI in Education 4.0
https://www3.weforum.org/docs/WEF_Shaping_the_Future_of_Learning_2024.pdf.
44. Osloo, S., (2023, 22 August). Why we must understand how generative AI will affect children. World Economic Forum.
https://www.weforum.org/agenda/2023/08/generative-ai-children-need-answers/.
45. Solyst, J., Yang, E., Xie, S., Hammer, J., Ogan, A., Eslami, M. (2024). Children’s Overtrust and Shifting Perspectives of
Generative AI. International Society of the Learning Sciences. https://arxiv.org/pdf/2404.14511.
46. Theil, D. (2023). Investigation finds AI image generation models trained on child abuse. Stanford Cyber Policy Center,
Stanford University. https://cyber.fsi.stanford.edu/news/investigation-finds-ai-image-generation-models-trained-child-
abuse.
47. Thiel, D., Melissa, S., Portnoff, R. (2023). New report finds generative machine learning exacerbates online sexual
exploitation. Stanford Digital Repository, Stanford University. https://cyber.fsi.stanford.edu/io/news/ml-csam-report.
48. World Economic Forum. (2023). Toolkit for Digital Safety Design Interventions and Innovations: Typology of Online Harms.
https://www3.weforum.org/docs/WEF_Typology_of_Online_Harms_2023.pdf.
49. Gruenhagen, J.H. et al. (2024). The rapid rise of generative AI and its implications for academic integrity: Students’
perceptions and use of chatbots for assistance with assessments. Computers and Education: Artificial Intelligence, vol. 7.
https://www.sciencedirect.com/science/article/pii/S2666920X24000766.
50. UNICEF. (n.d.). Policy guidance on AI for children. https://www.unicef.org/innocenti/reports/policy-guidance-ai-children.

51. Joint Research Centre, European Commission. (2022). Examining artificial intelligence technologies through the lens
of children’s rights. https://joint-research-centre.ec.europa.eu/jrc-news-and-updates/examining-artificial-intelligence-
technologies-through-lens-childrens-rights-2022-06-22_en.
52. World Economic Forum. (2022). Artificial Intelligence for Children: Toolkit.
https://www3.weforum.org/docs/WEF_Artificial_Intelligence_for_Children_2022.pdf.
53. Shekhawat, G., Livingstone, S. (2023). AI and children’s rights: A guide to the transnational guidance. London School
of Economics (LSE). https://blogs.lse.ac.uk/medialse/2023/11/01/ai-and-childrens-rights-a-guide-to-the-transnational-
guidance/.
54. Dotan, R. et al. (n.d.). Evaluating AI Governance: Insights from Public Disclosures. TechBetter.
https://www.techbetter.ai/_files/ugd/f83391_6aed42a5c87448b79821298183428a2e.pdf.
55. Li, F.-F. (2023). The Worlds I See: Curiosity, Exploration, and Discovery at the Dawn of AI. Flatiron Books.
56. Organisation for Economic Co-operation and Development. (2024). Governing with Artificial Intelligence:
Are governments ready? https://doi.org/https://doi.org/10.1787/26324bc2-en.
57. Li, F.-F. (2023). Governing AI Through Acquisition and Procurement. Stanford Institute for Human-Centered Artificial
Intelligence (HAI), Stanford University. https://hai.stanford.edu/sites/default/files/2023-09/Fei-Fei-Li-Senate-Testimony.pdf.
58. Cities for Digital Rights. (n.d.). Nine European cities set a common data algorithm register standard to promote
transparent AI. https://citiesfordigitalrights.org/9-european-cities-set-common-data-algorithm-register-standard-promote-
transparent-ai.
59. Australian Government Digital Transformation Agency. (2024). Policy for the Responsible Use of AI in Government.
https://www.digital.gov.au/sites/default/files/documents/2024-08/Policy%20for%20the%20responsible%20use%20
of%20AI%20in%20government%20v1.1.pdf.
60. The White House. (2024, 28 March). Fact sheet: Vice President Harris Announces OMB Policy to Advance Governance,
Innovation, and Risk Management in Federal Agencies’ Use of Artificial Intelligence [Press release].
https://www.whitehouse.gov/briefing-room/statements-releases/2024/03/28/fact-sheet-vice-president-harris-announces-
omb-policy-to-advance-governance-innovation-and-risk-management-in-federal-agencies-use-of-artificial-intelligence/.
61. Reda, M., Onsy, A., Haikal, A., Ghanbari, A. (2024). Path planning algorithms in the autonomous driving system:
A comprehensive review. Robotics and Autonomous Systems, vol. 174. https://doi.org/10.1016/j.robot.2024.104630.
62. Hambling, D. (2024). Hives For U.S. Drone Swarms Ready to Deploy This Year. Forbes.
https://www.forbes.com/sites/davidhambling/2024/05/16/hives-for-us-drone-swarms-ready-to-deploy-this-year/.
63. Heater, B. (2024). Figure’s new humanoid robot leverages OpenAI for natural speech conversations. TechCrunch.
https://techcrunch.com/2024/08/06/figures-new-humanoid-robot-leverages-openai-for-natural-speech-conversations/.
64. Interpol. (2024). Beyond Illusions: Unmasking the threat of synthetic media for law enforcement.
https://www.interpol.int/content/download/21179/file/BEYOND%20ILLUSIONS_Report_2024.pdf.
65. Associated Press. (2024). X restores Taylor Swift searches after deepfake explicit images triggered temporary block.
AP News. https://apnews.com/article/taylor-swift-x-searches-deepfake-images-adec3135afb1c6e5363c4e5dea1b7a72.
66. Thiel, D., Melissa, S., Portnoff, R. (2023). New report finds generative machine learning exacerbates online sexual
exploitation. Stanford Digital Repository, Stanford University. https://cyber.fsi.stanford.edu/io/news/ml-csam-report.
67. Shearer, J. (2024). Taylor Swift deepfakes on X falsely depict her supporting Trump. NBC News. NBC Universal.
https://www.nbcnews.com/tech/internet/taylor-swift-deepfake-x-falsely-depict-supporting-trump-grammys-flag-
rcna137620.
68. Martelloni, P.-H. (2021). Modélisation et Simulation des systèmes complexes spatialisés. Utilisation de Systèmes Multi-
Agents et Multi-composant pour la gestion des pêcheries. Universite de Corse-Pascal Paoli.
https://theses.hal.science/tel-03683015v1/document.
69. Samuel, S. (2024). People are falling in love with—And getting addicted to—AI voices. Vox.
https://www.vox.com/future-perfect/367188/love-addicted-ai-voice-human-gpt4-emotion.
70. Skaug Saetra, H., Mills, S. (2022). Psychological interference, liberty and technology. Technology in Society, vol. 69.
https://doi.org/https://doi.org/10.1016/j.techsoc.2022.101973.
71. Tong, A. (2023). AI chatbot company Replika restores erotic roleplay for some users. Reuters.
https://www.reuters.com/technology/ai-chatbot-company-replika-restores-erotic-roleplay-some-users-2023-03-25/.
72. Atillah, I. E. (2023). Man ends his life after an AI chatbot “encouraged” him to sacrifice himself to stop climate change.
Euro News. https://www.euronews.com/next/2023/03/31/man-ends-his-life-after-an-ai-chatbot-encouraged-him-to-
sacrifice-himself-to-stop-climate-.
73. World Economic Forum. (2024). Generative AI Governance: Shaping a Collective Global Future
https://www3.weforum.org/docs/WEF_Generative_AI_Governance_2024.pdf.
74. Taori, R., Hashimoto, T. B. (2022). Data Feedback Loops: Model-driven Amplification of Dataset Biases. Arxiv.
https://doi.org/10.48550/arXiv.2209.03942.
75. Shumailov, I. et al. (2024). The Curse of Recursion: Training on Generated Data Makes Models Forget. Arxiv.
https://doi.org/10.48550/ARXIV.2305.17493.

76. Finnish Government. (2023). Government Report on the Future.
https://valtioneuvosto.fi/en/foresight-activities-and-work-on-the-future/government-report-on-the-future.
77. Dubai Future Foundation. (n.d.). Foreseeing Dubai’s Future.
https://www.dubaifuture.ae/initiatives/future-foresight-and-imagination/dubai-future-councils/.
78. Lawfare, YouTube. (2024). Lawfare Daily: Former FCC Chair Tom Wheeler on AI Regulation.
https://www.youtube.com/watch?v=Oodn1zEjLvI.
79. Organisation for Economic Co-operation and Development (OECD). (2024). Regulatory Experimentation: Moving ahead
on the Agile Regulatory Governance Agenda. https://www.oecd.org/content/dam/oecd/en/publications/reports/2024/04/
regulatory-experimentation_fc84553c/f193910c-en.pdf.
80. Carnegie Endowment for International Peace. (2024). AI Governance for the Global Majority: Understanding Opportunities
and Challenges. https://carnegieendowment.org/events/2024/05/ai-governance-for-the-global-majority-understanding-
opportunities-and-challenges?lang=en.
81. World Economic Forum. (2024). Generative AI Governance: Shaping a Collective Global Future.
82. Alania, A. et al. (2022). Looking Ahead: The Role of Standards in the Future of Artificial Intelligence (AI) Governance.
University College London. https://www.ucl.ac.uk/steapp/sites/steapp/files/looking_ahead_the_role_of_standards_in_
the_future_of_ai_governance_v2.0.pdf.
83. Government of the United Kingdom. (2024, 21 May). Global leaders agree to launch first international network of AI Safety
Institutes to boost cooperation of AI [Press release]. https://www.gov.uk/government/news/global-leaders-agree-to-
launch-first-international-network-of-ai-safety-institutes-to-boost-understanding-of-ai.
84. World Economic Forum. (2024). Generative AI Governance: Shaping a Collective Global Future.
85. Carnegie Endowment for International Peace. (2024). AI Governance for the Global Majority: Understanding Opportunities
and Challenges. https://carnegieendowment.org/events/2024/05/ai-governance-for-the-global-majority-understanding-
opportunities-and-challenges?lang.

91–93 route de la Capite
CH-1223 Cologny/Geneva
Switzerland
Tel.: +41 (0) 22 869 1212
Fax: +41 (0) 22 786 2744
contact@weforum.org
www.weforum.org
The World Economic Forum,
committed to improving
the state of the world, is the
International Organization for
Public-Private Cooperation.
The Forum engages the
foremost political, business
and other leaders of society
to shape global, regional
and industry agendas.

Governance in the Age of Generative AI: A 360º Approach for Resilient Policy and Regulation

More Related Content

Similar to Governance in the Age of Generative AI: A 360º Approach for Resilient Policy and Regulation (20)

More from Willy Marroquin (WillyDevNET) (20)

Recently uploaded (20)

Governance in the Age of Generative AI: A 360º Approach for Resilient Policy and Regulation