First International AI Safety Report is published ahead of the Paris summit
A Bengio-chaired, 30-country-backed synthesis of AI capability and risk research became the first government-commissioned cross-national scientific consensus document.
- Ideas & essays
- Government & policy
- Safety & alignment
- Major
The first full International AI Safety Report was published, a synthesis of research on the capabilities and risks of general-purpose AI systems chaired by Yoshua Bengio and written with around 100 experts nominated by an Expert Advisory Panel drawn from 30 countries plus the UN, OECD and EU. It had been commissioned at the November 2023 AI Safety Summit at Bletchley Park, explicitly modelled on the UN’s Intergovernmental Panel on Climate Change process, and was intended to give governments a shared evidence base ahead of the AI Action Summit in Paris the following month. An interim version had already been produced for the May 2024 Seoul summit.
The report grouped risks into three categories: malicious use, such as cyberattacks, deepfakes and biosecurity threats; malfunctions, including bias, unreliability and loss of control; and systemic risks, covering labour-market disruption, environmental cost and market concentration among a small number of developers. Rather than issuing its own risk assessment or policy recommendations, it aimed to state where the evidence was strong, where it was contested, and where it was simply too thin to support confident claims — a deliberate contrast to advocacy documents on either side of the AI-risk debate.
As a government-commissioned, multi-country scientific consensus exercise on AI, it had no real precedent; previous influential documents on AI risk had come from individual labs, academics or civil-society coalitions rather than a process with formal government sponsorship across dozens of countries. Its neutral framing was also a limitation: because it deliberately avoided taking a position on contested questions — how likely near-term loss-of-control risks were, for instance — it was more useful as a shared reference than as a guide to what any government should do next.
The report was presented at the Paris summit, where the political mood had shifted markedly from the safety-focused tone of Bletchley; a second edition followed in 2026.