To better understand & navigate the path to superintelligence & prevent the creation of uncontrolled AI, we must build trust across corporations & countries, create safety & verification mechanisms, avoid dangerous power imbalance, and ensure AI's benefits are shared globally. campaign.controlai.com

My paper with @ericelmoznino.bsky.social on AI consciousness and the danger of people seeing it (whether or not it is there) is out in @science.org. www.science.org/doi/10.1126/...

I recently discussed my work on building honest AI as a guardrail for the risks of frontier AI systems w/ Pier Luigi Pisa of La Repubblica. We also covered the potential economic inequities the global AI race could cause & how it could put Europe at a disadvantage. www.repubblica.it/tecnologia/2...

This text in Corriere della Sera is a great recap of the rich discussions I’ve had in Italy this week with the AI & Fraternity Working Group coordinated by @riccardoluna.bsky.social. The Vatican is playing a crucial role in the global dialogue on AI. www.corriere.it/tecnologia/2...

I'm currently in Rome as part of the AI & Fraternity Working Group with a small group of AI experts invited by the Fratelli Tutti Foundation. Very grateful to be able to reflect on the role & risks of AI for humanity, and for the chance to discuss with Fr. Paolo Benanti in person.

Félicitations et merci Hugo de reprendre le flambeau en tant que directeur scientifique de @mila-quebec.bsky.social. Ta rigueur, ta créativité et ta vision seront des atouts essentiels pour les prochains chapitres de notre institut.

Grateful to be recognized on TIME’s list of the TIME100 AI of 2025! #TIME100AI time.com/collections/...

Global cooperation & coordination are among the most urgent tasks to mitigate AI risks & ensure the technology benefits all. Congratulations to the United Nations on establishing the Independent International Scientific Panel on AI & the Global Dialogue on AI Governance. media.un.org/unifeed/en/a...

This work will be instrumental in supporting the creation of our technical solution to address critical AI safety challenges and propel scientific discovery. Thank you to the Gates Foundation!

Planning capabilities double every 7mo→human-level in 5yrs? @yoshuabengio.bsky.social: "We still don't know how to make sure powerful AIs won't turn against us" AIs now lie to avoid shutdown, self-preserve. Solution: Non-agentic "Scientist AIs" + global governance beyond market forces 👇

It was a pleasure to meet Prime Minister @mark-carney.bsky.social and Minister Evan Solomon today as part of an official visit at @mila-quebec.bsky.social with many collaborators from our AI ecosystem. We discussed AI risk mitigation, sovereignty & economic potential among other important topics.

This video is a great illustration of scenarios from Daniel Kokotajlo et al’s “AI 2027,” highlighting the major risks of the race toward AGI. AI development needs to be steered towards safer, more beneficial outcomes. www.youtube.com/watch?v=5KVD...

Very insightful video by Tristan Harris on the growing risks of AI and the race to develop it—but just as importantly, on the danger of thinking our current path is inevitable. We still have the chance to coordinate globally and create a safer way forward. youtube.com/watch?v=86k8N4YsA7c

I’ve been thrilled to see the support for the Safety & Security Chapter of the Code of Practice. Most frontier AI companies have now signed on to it: @anthropic.com, Google, Mistral AI, OpenAI, xAI. Why this is important: 🧵 1/7

The Economist offers an accurate overview of key forces shaping the development of AI today: the risks of the race toward AGI and ASI, the challenges posed by open-sourcing frontier models, the deep uncertainty revealed by ongoing scientific debates and much more ⬇️ www.economist.com/briefing/202...

Delighted to be working with you again and embarking on this new adventure, Philippe. Welcome to the team!

Ravi de te retrouver et de commencer cette nouvelle aventure avec toi Philippe. Bienvenue dans l’équipe!

The future of AI governance may hinge on our ability to develop trusted and effective ways to make credible claims about AI systems. This new report expands our understanding of the verification challenge and maps out compelling areas for further work. ⬇️

The Code of Practice is out. I co-wrote the Safety & Security Chapter, which is an implementation tool to help frontier AI companies comply with the EU AI Act in a lean but effective way. I am proud of the result! 1/3

By advancing SB 53, California is uniquely positioned to continue supporting cutting-edge AI while proactively taking a step towards addressing the severe and potentially irreversible harms that frontier systems could cause.

Excited to share our paper: "Chain-of-Thought Is Not Explainability"! We unpack a critical misconception in AI: models explaining their steps (CoT) aren't necessarily revealing their true reasoning. Spoiler: the transparency can be an illusion. (1/9) 🧵

📢 £18m grant opportunity: in Safeguarded AI we're looking to catalyse the creation of a new UK-based non-profit to lead groundbreaking machine learning research for provably safe AI. Learn more and apply by 1 October 2025: link.aria.org.uk/ta2-phase2-bs

The timeline & severity of major AI risks are still debated within the scientific community, but these disagreements reveal great uncertainty. The fact that many credible experts consider some catastrophic scenarios plausible should be enough to warrant serious caution. www.axios.com/2025/06/16/a...

Enjoyed speaking with @sigalsamuel.bsky.social of @vox.com to mark the launch of @law-zero.bsky.social. We discussed the motivation behind the project, its research direction, and the challenges and risks of increasingly capable and autonomous AI systems. Full article: www.vox.com/future-perfe...

Welcome to the team Sam! Your leadership and expertise will be huge assets for @law-zero.bsky.social moving forward.

This report from the Joint California Policy Working Group on AI Frontier Models is an important step towards effective and balanced AI regulation. It builds meaningfully on the International AI Safety Report and offers a thoughtful framework for policymaking at the frontier of AI development. 🧵 1/4

The @ec.europa.eu is setting up a scientific panel of 60 independent experts to support the implementation and enforcement of the EU AI Act. This is a great opportunity to contribute to AI governance in Europe. More details below ⤵️ digital-strategy.ec.europa.eu/en/news/comm...

I recently spoke with BBC World Service's James Copnall about early signs of deception and self-preservation in frontier AI systems, and shared key ideas behind Scientist AI, the focus of @law-zero.bsky.social's research. 🎧 Listen to the full interview here: www.bbc.co.uk/programmes/w...

Je suis très honoré de cette nomination à titre d’officier de l’Ordre national du Québec; cette nation où j’ai grandi, étudié, enseigné, et qui, il y a une dizaine d’années, a été l'une des premières à soutenir développement responsable & éthique de l'IA www.ordre-national.gouv.qc.ca/actualites/n...

I was recently on @bbcnewsnight.bsky.social to discuss the accelerating pace of advances in AI capabilities & the emergence of deceptive & self-preserving behaviours in frontier models. The conversation covers many issues that led me to launch @law-zero.bsky.social. www.youtube.com/watch?v=c4Zx...

Il s’agit d’un grand jour pour moi. Je lance LoiZéro @law-zero.bsky.social, un OBNL qui se penchera sur une nouvelle approche de l’IA plus sécuritaire qui pourrait permettre d'accélérer la découverte scientifique et d'éviter les dérives liées à l’IA agentique.

Today marks a big milestone for me. I'm launching @law-zero.bsky.social, a nonprofit focusing on a new safe-by-design approach to AI that could both accelerate scientific discovery and provide a safeguard against the dangers of agentic AI.

I recently participated in the High-Level Expert Panel on AGI, convened by the Council of Presidents of the UN General Assembly, which has just released its final report on the Governance of the Transition to AGI. Report below ⬇️ uncpga.world/agi-uncpga-r...

When I realized how dangerous the current agency-driven AI trajectory could be for future generations, I knew I had to do all I could to make AI safer. I recently shared this personal experience, and outlined the scientific solution I envision @TEDTalks⤵️ www.ted.com/talks/yoshua...

In the absence of a robust federal regulatory framework, state-level oversight remains a critical societal line of defense against the risks posed by the rapid advancement of frontier AI. Preempting would, in effect, fail to protect the public. thehill.com/policy/techn...

Thank you to @politico.eu and @pieterhaeck.bsky.social for a great conversation on how the latest advances in AI are shaping Europe’s present and future. 📽️Full discussion available here: www.linkedin.com/posts/politi...

Two years ago, I've completely reoriented my research to try to make AI safe by design. In this TIME op-ed, I present my team's new direction called "Scientist AI"; a practical, effective and more secure alternative to the current uncontrolled agency-driven trajectory. time.com/7283507/safe...

This gathering demonstrated global convergence around the technical challenges in AI safety. By facilitating exchanges between scientists and policymakers, we can work to build a trusted ecosystem that delivers maximum innovation while harnessing AI for the public good. aisafetypriorities.org

Rival nations or companies sometimes choose to cooperate because some areas are protected zones of mutual interest—reducing shared risks without giving competitors an edge. Our paper in FAccT '25: How geopolitical rivals can cooperate on AI safety research. arxiv.org/abs/2504.12914

This metric from METR shows the length of tasks AI agents can complete has been consistently exponentially increasing over the past 6 years, with a doubling time of around 7 months. metr.org/blog/2025-03... 1/2

Over 100 researchers are issuing a statement today that reiterates the scientific consensus that AI can exacerbate bias and discrimination. The last three U.S. Administrations — including the first Trump Administration — and 192 other nations have recognized this basic fact.

Very relevant piece by @kevinroose.com in @nytimes.com, the right time to start preparing for AGI is now. www.nytimes.com/2025/03/14/t... 3 points that particularly resonate with me, 🧵⬇️ 1/3

We welcome continued feedback and discussion as we head into the final stages of drafting.

The 3rd draft of the EU's GPAI Code of Practice has just been published. It provides clarity on a way to comply with the EU AI Act. I've been co-chairing the Safety & Security Section, focused on managing risks at the frontier of AI development. Full text: code-of-practice.ai?section=safe...

In this interview with Stephanie Hare for BBC, I shared my recent thoughts on the progression of models in programming and reasoning, and especially on how the next frontier, agency, will force us to quickly develop technical and societal safeguards. www.youtube.com/watch?v=3cFk...

Je recommande vivement la pièce Une vie intelligente de Dominique Leclerc qui est présentée au @theatre-duceppe.bsky.social jusqu'au 29 mars. Il s'agit d'un produit culturel très pertinent et utile pour comprendre l'ampleur des répercussions de l'IA sur nos vies. duceppe.com/une-vie-inte...

Le 28 février, j'ai eu l'honneur de recevoir la Médaille du couronnement du roi Charles III en reconnaissance de mes contributions en apprentissage profond. Je remercie le Sénateur Andrew Cardozo de m'avoir remis cette importante distinction!

On Feb. 28, I was honoured to receive the King Charles III Coronation Medal in recognition of my contributions to deep learning. Thank you to Senator Andrew Cardozo for presenting me with this important distinction!

Congratulations to Rich Sutton and Andrew Barto on receiving the Turing Award in recognition of their significant contributions to ML. I also stand with them: Releasing models to the public without the right technical and societal safeguards is irresponsible. www.ft.com/content/d8f8...

Early signs of deception, cheating & self-preservation in top-performing models in terms of reasoning are extremely worrisome. We don't know how to guarantee AI won't have undesired behavior to reach goals & this must be addressed before deploying powerful autonomous agents. time.com/7259395/ai-c...