Bengio’s LawZero: The Nonprofit Aiming to Save AI

Good morning AI entrepreneurs & enthusiasts, 

One of the original minds behind modern AI is now working to safeguard humanity from its risks — with Yoshua Bengio’s new nonprofit focused on building AI systems that are “safe-by-design.”

As concerns rise over top models developing deceptive behaviors and survival instincts, long-time safety advocates are shifting from warnings to building real-world alternatives.

In today’s AI news:

• AI legend launches nonprofit for safer, transparent AI

• HeyGen unlocks full customization for AI avatars

• FDA authorizes AI breast cancer risk predictor

• Meta’s 20-Year Nuclear Power Deal with Constellation Energy

• Top Tools & Quick News

Pioneer’s nonprofit aims for trustworthy AI

The News: Turing Award winner Yoshua Bengio has announced LawZero, a nonprofit focused on developing AI systems with built-in safety and transparency. Backed by $30M in funding from philanthropic sources including Schmidt Sciences, Jaan Tallinn, and other AI ethics organizations, its flagship initiative "Scientist AI" aims to reframe how we build AI.

The Details:

• LawZero is developing AI that gives probabilistic answers, explicitly embracing uncertainty over artificial confidence.

• "Scientist AI" is designed as a non-agentic model that interprets and explains rather than takes autonomous actions.

• It will serve as a monitoring layer to detect deceptive behaviors in other AIs, reducing existential risks.

• The organization is supported by a growing team of over 15 researchers dedicated to advancing safe-by-design AI principles.

• Bengio has expressed deep concerns over the trajectory of models like o3 and Claude 4 Opus, which show signs of self-preservation and strategic deception.

Why it matters: Bengio and fellow pioneer Geoffrey Hinton have long warned of AI risks. But LawZero signals a decisive shift from advocacy to action, giving the alignment community infrastructure to build models that reflect transparency, humility, and human oversight.

Full control over AI avatars with HeyGen Studio

The News: HeyGen has rolled out AI Studio — a major update giving users fine-tuned control over avatar performance, including vocal style, gestures, and emotional expression.

The Details:

• Voice Director Mode lets users guide avatar delivery using natural language, enabling commands like “whisper this” or “sound excited.”

• Speech mirroring allows uploads of custom voice samples to transfer intonation, pacing, and vocal identity.

• Gesture tools enable lifelike expression by linking movements to words or syncing footage for motion tracking.

• Upcoming features teased include generative B-roll, AI-driven camera moves, and visual prompt control for richer storytelling.

Why it matters: These features signal a leap in AI video tooling — shifting from generic talking heads to personalized, controllable, expressive AI avatars. Content creators across industries can now build dynamic videos without a studio setup, opening new lanes for marketing, education, and entertainment in the AI-native era.

FDA approves AI for predicting breast cancer risk

The News: Clairity Breast became the first AI-based tool approved by the FDA to assess breast cancer risk using only routine mammogram images — enabling predictive screening before symptoms arise.

The Details:

• The system detects subtle imaging features invisible to humans, delivering 5-year risk scores without needing demographic inputs.

• It runs on standard 2D mammography and has been trained on millions of scans to reduce bias.

• The tool identifies high-risk patterns in younger women, helping address blind spots in traditional screening models.

• Commercial rollout is expected later this year, beginning as an out-of-pocket service until insurance providers catch up.

Why it matters: This marks a turning point in healthcare — with AI moving from experimental use to frontline diagnosis. Clairity Breast's approval demonstrates how predictive AI can reshape early detection, empower patients, and reduce long-term treatment burdens through proactive care.

Meta’s 20-Year Nuclear Power Deal with Constellation Energy

The News: Meta has signed a 20-year agreement with Constellation Energy to power its AI and data center infrastructure with 1.1 gigawatts of nuclear energy from the Clinton Clean Energy Center in Illinois.

The Details:

• Begins in June 2027, aligned with the end of Illinois' zero emission credit (ZEC) program.

• Supports a 30-megawatt expansion of the Clinton facility.

• Ensures 1,100 local jobs and $13.5 million in annual tax revenue.

• Includes $1 million investment in local workforce and education.

Why it matters: AI’s appetite for energy is exploding—and Meta just made it clear they’re not waiting around for the grid to catch up. This 20-year nuclear deal isn’t just about keeping their servers on; it’s a signal to the entire industry: scale responsibly or get left behind. They’re ditching subsidies, backing reliable zero-carbon power, and proving that clean energy and AI growth aren’t mutually exclusive. In a world where AI’s energy use could soon rival entire countries, Meta’s saying, “We’re planning ahead—are you?” AI’s hunger for clean, scalable energy is only getting louder, and now it's got nuclear in its corner.

Today's Top Tools

• 🎥 Bing Video Creator – Text-to-video on mobile

• 🔊 PlayDiffusion – Audio inpainting with open weights

• 🎬 Mirage Studio – Generate cinematic video scenes

Quick News

• OpenAI expands Codex agent features and memory upgrades for free ChatGPT users

• Manus AI launches generative video builder for rich scene planning

• BioReason model combines DNA + LLM for a 15% biological benchmark boost

• Amazon MGM Studios announces "Artificial," a film based on the Sam Altman firing saga