cybersecurity-for-ai-and-genai-updated.pdf

Cybersecurity meets
AI and GenAI
October 2024

© 2024. For information, contact Deloitte Global.
Table of contents
Dimensions of Cybersecurity within AI and GenAI
1 03
2 AI and GenAI induced change in Cybersecurity threat landscape 04-05
Cybersecurity for AI and GenAI Framework legend
5 08
4 AI and GenAI Cybersecurity roadmap – discover your next steps 07
3 Cybersecurity for AI and GenAI Framework 06

Cybersecurity considerations with regards to Artificial Intelligence (AI) and Generative AI (GenAI) have to be
viewed from three different angles: Securing AI and GenAI Systems, using AI and GenAI for improving
Cybersecurity, and using AI and GenAI for malicious actions.
Dimensions of Cybersecurity within AI and GenAI
1
3
Cybersecurity
For AI and GenAI
Protecting AI and GenAI
systems from Cybersecurity
threats, by providing
guidance to secure
implemented or planned AI
and GenAI use-cases.
Cybersecurity
With AI and GenAI
Improving Cybersecurity
capabilities and boosting
Cybersecurity processes
by including AI and GenAI.
Cybersecurity
From AI and GenAI
Changing Cybersecurity
threat landscape due to
launch of more sophisticated
and new kinds of
cyberattacks.
Use Case Ideation &
Development
AI and GenAI Training
& Labs
C y b e r s e c u r i t y
F o r A I a n d G e n A I
C y b e r -
s e c u r i t y
W i t h
A I a n d
G e n A I
C y b e r -
s e c u r i t y
F r o m
A I a n d G e n A I
Cybersecurity for AI
and GenAI
Framework
Trusted & Secure AI
Focus of the following
“Futurecasting” AI and
GenAI Tabletop
Exercises
AI Threat Intel & Attack
Surface Management

The rise of AI and GenAI not only comes with new opportunities but also with a change in security-related
threats that will continue to evolve, making it imperative to secure AI systems.
AI and GenAI induced change in Cybersecurity threat landscape*
1
4
M O D E L
Attacker
Cybersecurity Threat Landscape
The increasing usage and availability of AI and
GenAI leads to a change in the Cybersecurity
threat landscape. On the one hand it enables
attacker to intensify their attack frequency,
efficiency and complexity due to the use of AI and
GenAI, on the other hand it is leading to
completely new threats for AI and GenAI like
adversarial attacks.
Moreover, the attack surface presented by AI and
GenAI solutions is unfamiliar territory for many. It’s
not only the infrastructure, data and application
that require safeguarding, but also the underlying
model on which any AI and GenAI System is build.
It contains many sensitive information and
requires additional protection. Additionally, the
increasing amount of data being processed and
stored is leading to an increasing focus of data
security.
A p p l i c a t i o n
M o d e l
I n f r a s t r u c t u r e
D a t a
*based on ISO standard (ISO/SAE42001:2023), OWASP (threats for LLM and ML) and ENISA publications

Based on the publications of OWASP, ISO and ENISA, Deloitte consolidated the Top 10 threats for AI and
GenAI.
AI and GenAI is expanding the Cybersecurity threat landscape*
1
5
Excessive agency abuse
Causing AI applications to gain excessive authority or
using such authority to perform unintended actions
beyond desired functionality.
Model inversion
Utilizing an AI model's output to reconstruct sensitive
data samples used for training, effectively reverse-
engineering the model to extract information.
Model poisoning
directly manipulating AI model's parameters to
influence its behavior negatively.
Model stealing
Unauthorized access, copying, or exfiltration of an AI
model.
Input Injection
Compromising AI applications with malicious inputs
that override controls or alter model behavior e.g.
Prompt injection for Large Language Models (LLMs).
Training data poisoning
Introducing vulnerabilities or biases into AI models
by tampering their training data, compromising
security, effectiveness, or ethical behavior.
Information breach
Unauthorized exposure of private data and/or
metadata leading to unwarranted data access,
privacy (GDPR) violations, and security breaches.
Supply chain vulnerabilities
Incorporating compromised or insecure third-party
components like third-party datasets, pre-trained
models, and plugins leading to security risks.
Adversarial examples
Utilizing adversarial learning to create malicious
inputs, which deceive AI models during the inference
phase, e.g. by causing a misclassification.
Model denial of service
Triggering resource-intensive operations through
inputs that lead to AI model disruptions.
Current
AI and GenAI
Threats

Overarching & AI and GenAI domain specific security capabilities ensure the secure development,
implementation, and usage of AI and GenAI solutions.
Cybersecurity for AI and GenAI Framework*
1
6
Lifecycle Security
Secure supply chain
Governance, Risk & Compliance
The domains constitute the core structure of
AI and GenAI systems and are used to
cluster security capabilities
The Data Domain includes all data handled
by the model during training, testing,
validation, and for inference after
deployment.
The AI and GenAI Model Domain involves the
model architecture, training, testing and
validation processes, in addition to the
model’s unique parameters.
The Application Domain is the external layer
of the AI and GenAI system that hosts the
model and sits on the infrastructure. It
serves as the user interface.
The Infrastructure Domain encompasses the
underlying hardware and networking
components that are used for developing
and hosting the AI and GenAI system.
Data Integrity
& Quality
Data Lineage &
Provenance
Model Security
Testing
Adversarial
Machine Learning
Model Access
Controls
User Access
Controls
Application Logging &
Monitoring
Model Behavior
Monitoring
Application
Architecture Sec.
Network
Security
Infrastructure
Security
Cloud
Security
Secure Model
Development
User Abuse
Monitoring
Data Privacy &
Privacy Enhancing
Technologies
A Security Capability is a category for
grouping of controls that are designed to
help address specific Cybersecurity
threats in each domain.
Capabilities to maintain the security of AI and
GenAI solutions across each domains.
Secure development process
Data loss prevention (DLP)
Asset management
AI and GenAI Domains
AI and GenAI Security Capabilities
Overarching Security Capabilities
Regulatory compliance
Third-party risk management
AI and GenAI security risk management
AI and GenAI specific policies, standards & architecture
Business continuity management
D
a
t a
A I a n d
G
e
n
A
I
M
o
d
e
l
A p p l i c
a
t
i
o
n
I
n
f
r
a
s
t
r
u
c t u r e

The AI and GenAI Cybersecurity Roadmap is designed to help organizations on the journey toward secure
implementation, deployment, and usage of AI and GenAI applications.
AI and GenAI Cybersecurity roadmap – discover your next steps
1
7
Understand the basics: delve into foundational
concepts of Cybersecurity for AI and GenAI including
threat landscape, encryption, network security, and
access controls with AI labs and future casting table-
top exercises (TTX)
Familiarize yourself with AI and GenAI: gain a basic
understanding of AI and GenAI principles and its
implementation, algorithms, and its applications in
Cybersecurity.
Hold a AI and GenAI
Cybersecurity lab
01 03
02 04
05
Assess your AI risk level
(AIRL)
Measure your maturity/risk level: to gauge
your organization's readiness and maturity, we
have Framework devised a broad assessment and
security. Deloitte's solution can help you define the
AIRL of each component you are hoping to secure
scoring it on a 1 to 5 scale. Your components’ AIRL
will inform the specific controls families to be
considered and prioritized.
Identify tailored
Security controls
Identify the tailored set of Cybersecurity
controls. Deloitte’s Cybersecurity for AI and
GenAI Framework has over 500 controls from
legislation, industry standards, and existing
frameworks mapped to four domains to meet
the needs of your AIRL.
Implement tailored
Security controls
Begin with individual quick wins: Deloitte’s
approach begins with implementing individual
quick wins derived from our maturity
assessment to focus on near-term security
enhancements. These quick wins are
actionable steps that are designed to help you
yield significant improvements in your security
posture.
Actively monitor your AI and GenAI systems.
Though implementing tailored security controls
is crucial to harden your AI and GenAI systems
against threats, the ever-evolving threat
landscape and new legislation/regulations are
pushing organizations to continuously monitor
their environments to stay abreast of attacks.
To bolster your confidence in the security of
your AI systems, Deloitte’s attack surface
monitoring (ASM) and threat intel, and AI red
teaming services are next steps to help you
bolster the Cybersecurity for and from AI
systems and threat actors.
Engage in continuous
monitoring

The Deloitte Cybersecurity for AI and GenAI Framework synthesizes best practices from ISO standards, MITRE
ATT&CK, OWASP and ENISA to secure your AI and GenAI system to combat known and novel threats.
Cybersecurity for AI and GenAI Framework legend
1
8
Controls
For each of the capabilities, security
controls are summarized to mitigate
all potential threats.
Extract of controls
• Robust access control
• Regular validation of SBOM
• Watermarking
• Input injection testing
• Data Minimization
• Data Separation
System domains
AI and GenAI threats identified can be exploited
in different stages of the AI and GenAI lifecycle.
Building appropriate mitigation mechanisms against
the new and traditional vulnerabilities, the AI and
GenAI system is organized in four different domains.
Cybersecurity for AI and GenAI Framework structure
Threats
As a special type of software,
there are new and traditional
Cybersecurity threats to AI and
GenAI systems.
Regulations
We have conducted geography-
specific research to comply with
regulations and follow
standards.
This is a major focal point of this
Framework and is being
continuously updated to keep
up with this evolving space.
Domain-specific capabilities
For each domain, there are different
capabilities which group security
controls to address all the potential
attack vectors and threats for AI and
GenAI.
Overarching capabilities
As AI and GenAI is a software at its core,
overarching capabilities ensures
Governance, Risk, and Compliance for
the AI and GenAI Lifecycle across
domains.
External drivers
D a t a
All data used in training,
testing, validation,
and
post-deployment
inference.
A I a n d G e n A I
M o d e l
The model’s
architecture, its
training, testing,
validation, and
parameters.
Hardware and
networking
components for
developing and
hosting the system.
I n f r a s t r u c t u r e
The external
layer hosting the
model, acting as
the user interface.
A p p l i c a t i o n
Stay tuned for our next publication!
Lifecycle Security
Secure supplychain
Secure developmentprocess
Data loss prevention (DLP)
Asset management
Governance, Risk & Compliance
Regulatory compliance
Third-party risk management
AI and GenAI securityrisk management
AI and GenAI specificpolicies,standards & architecture
Business continuitymanagement

Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited
(DTTL), its global network of member firms, and their related entities
(collectively, the “Deloitte organization”). DTTL (also referred to as
“Deloitte Global”) and each of its member firms and related entities
are legally separate and independent entities, which cannot obligate
or bind each other in respect of third parties. DTTL and each DTTL
member firm and related entity is liable only for its own acts and
omissions, and not those of each other. DTTL does not provide
services to clients. Please see www.deloitte.com/about to learn more.
Deloitte provides industry-leading audit and assurance, tax and legal,
consulting, financial advisory, and risk advisory services to nearly 90%
of the Fortune Global 500® and thousands of private companies. Our
people deliver measurable and lasting results that help reinforce
public trust in capital markets, enable clients to transform and thrive,
and lead the way toward a stronger economy, a more equitable
society, and a sustainable world. Building on its 175-plus year history,
Deloitte spans more than 150 countries and territories. Learn how
Deloitte’s approximately 457,000 people worldwide make an impact
that matters at www.deloitte.com.
This communication contains general information only, and none of
Deloitte Touche Tohmatsu Limited (DTTL), its global network of
member firms or their related entities (collectively, the “Deloitte
organization”) is, by means of this communication, rendering
professional advice or services. Before making any decision or taking
any action that may affect your finances or your business, you should
consult a qualified professional adviser.
No representations, warranties or undertakings (express or implied)
are given as to the accuracy or completeness of the information in
this communication, and none of DTTL, its member firms, related identities,
employees or agents shall be liable or responsible for any
loss or damage whatsoever arising directly or indirectly in connection
with any person relying on this communication. DTTL and each of its
member firms, and their related entities, are legally separate and
independent entities.
© 2024. For information, contact Deloitte Global
Thank you.
Volker Burgers
Partner
vburgers@deloitte.de
Tim LI
Principal
timli@deloitte.com
Jordan McKenzie
Manager
jormckenzie@deloitte.de
Lucie Wollenhaupt
Manager
lwollenhaupt@deloitte.de

cybersecurity-for-ai-and-genai-updated.pdf

More Related Content

Similar to cybersecurity-for-ai-and-genai-updated.pdf (20)

Recently uploaded (20)

cybersecurity-for-ai-and-genai-updated.pdf