Yoshua Bengio
Deep-learning pioneer; 2018 Turing Award; chair of the International AI Safety Report; founder of LawZero
Université de Montréal / Mila / LawZero
Bengio is one of the scientists who built modern AI and now leads the world's main government-backed scientific assessment of AI risk. He has said there is roughly a one-in-five chance things turn out catastrophic, and in 2025 founded a nonprofit to build AI that understands the world without acting in it. He wants binding international rules and has signed calls to pause superintelligence development.
Their number
20%
"I got around, like, 20 per cent probability that it turns out catastrophic." (ABC News Australia, Background Briefing, July 15, 2023). He added: "Even if it was 0.1 per cent [chance of doom], I would be worried enough to say I'm going to devote the rest of my life to trying to prevent that from happening." In December 2025 he declined to give a single personal figure, citing surveys of ML researchers 'more like 10%' and calling even 1% 'unacceptable'.Source
On timing
"My current estimate places a 95% confidence interval for the time horizon of superhuman intelligence at 5 to 20 years" (FAQ on Catastrophic AI Risks, June 24, 2023). In December 2025 (Diary of a CEO) he said threatening systems "could be in a few years or a decade or two."
SourceWhat they call for
- International treaties 'with real teeth' and mutual US China verification
- Licensing and monitoring of frontier AI
- No open sourcing of the most powerful generalist models
- Mandatory liability insurance for developers
- Non agentic 'Scientist AI' as a safer design
- International agreement on AI red lines
- Prohibition on superintelligence until safety consensus
In their own words
Launched LawZero "in response to evidence that today's frontier AI models have growing dangerous capabilities and behaviours, including deception, cheating, lying, hacking, self-preservation, and more generally, goal misalignment."
3 June 2025
Chaired the International AI Safety Report 2026 (100+ experts, 30+ countries), which found current systems 'lack the capabilities' for loss-of-control risks 'but they are improving in relevant areas such as autonomous operation' and that 'technical safeguards are improving but still show significant limitations.'
3 February 2026
Signed the Statement on Superintelligence calling for 'a prohibition on the development of superintelligence' until there is scientific consensus on safety and public buy-in.
22 October 2025
Signed
- FLI pause letter 2023
- CAIS extinction statement 2023
- Right to Warn letter 2024 (endorser)
- Global Call for AI Red Lines 2025
- FLI Statement on Superintelligence 2025