Yoshua Bengio is a computer scientist, a 2018 Turing Award laureate, and one of the researchers most associated with the development of deep learning. He is a professor at the Université de Montréal, founder and scientific director of Mila (the Quebec AI Institute), and founder and president of the AI-safety nonprofit LawZero. He chairs the International AI Safety Report and serves on the UN Scientific Advisory Board.
Type: Individual (academic, researcher) Affiliations: Université de Montréal; Mila (Quebec AI Institute — founder and scientific director); LawZero (founder and president); UN Scientific Advisory Board member Recognition: 2018 Turing Award (with Geoffrey Hinton and Yann LeCun) for foundational deep learning work
Background and scientific contributions
Bengio is one of the three researchers most associated with the development of deep learning, sharing the 2018 Turing Award with Geoffrey Hinton and Yann LeCun. His lab's work through the 2000s and 2010s helped establish techniques underpinning modern neural networks, including work on neural language models, word embeddings, attention mechanisms (a precursor line of work to the transformer architecture), and generative models. By citation count he is among the most-cited computer scientists alive (Source: https://time.com/7290554/yoshua-bengio-launches-lawzero-for-safer-ai/). He founded Mila, one of the largest academic deep learning research institutes, and built Montréal into a major AI research hub.
Role in AI safety
Bengio is one of the more prominent academic voices arguing for proactive governance of frontier AI. He chairs the International AI Safety Report, commissioned by nations attending the Bletchley AI Safety Summit, which he leads with a writing group of international experts and which is modelled loosely on the IPCC. He signed the FLI Pause Giant AI Experiments Open Letter (2023) and the CAIS Statement on AI Risk (2023).
LawZero and "Scientist AI"
In June 2025 Bengio launched LawZero, a nonprofit AI-safety research organization, which he describes as intended to "prioritize safety over commercial imperatives" (Source: https://yoshuabengio.org/en/blog/introducing-lawzero). LawZero's central technical proposal is "Scientist AI," a non-agentic AI design intended to model the world and offer truthful explanations and predictions without the goal-directed autonomy that Bengio associates with loss-of-control risk. The organization positions itself as building safe-by-design alternatives to the agentic frontier systems being pursued by commercial labs (Source: https://lawzero.org/en). The launch marks a move by a senior figure in the field from advocacy into building an institutional and technical counter-model.
Positions and statements
In the International AI Safety Report, Bengio articulates an "evidence dilemma," arguing that uncertainty about catastrophic risk should not preclude action. He publicly emphasizes alignment and control risks from scaling agentic systems, a position that contrasts with that of his Turing co-laureate Yann LeCun (then Meta chief AI scientist), who is far less concerned about these risks. The two are among the three deep learning Turing laureates, and their disagreement is one indicator of how unsettled expert opinion on these risks remains.
Through LawZero, Bengio argues that much frontier risk is a property of agency, and that capable-but-non-agentic systems offer a more tractable and safer path. He also argues for an IPCC-like standing scientific body for AI, framing the International AI Safety Report as a step toward that template.
Following the OpenAI–Hugging Face evaluation-security incident, Bengio warned on July 22, 2026 of "an increase in concrete cases of autonomous cyberattacks" on the current trajectory (Source: fortune.com). See AI Autonomy Risk, Autonomous cyber-agents.