AI experts share a paper ahead of the UK's AI summit in Bletchley Park warning of worrying capabilities in current AI systems and seeking urgent governance
Authors and academics also warn development of advanced systems ‘utterly reckless’ without safety checks
The GuardianDan Milmo
Context & Ripple Effects
The intervention arrived as the UK was convening a summit around advanced AI risks, extending a longer policy debate that had already mapped malicious uses of AI across digital, physical and political domains.
It also exposed a gap between safety rhetoric and implementation: a prior critique said the UK's safety approach lacked credibility, while later coverage showed the debate reaching whether to pause next-generation frontier models.
First-order effects
The paper gives UK summit participants a concrete expert-backed case for urgent governance and safety checks around advanced systems.
It raises immediate pressure on AI developers and funders to treat risk management as a budgeted operating responsibility rather than a voluntary statement of principle.
Second-order effects
Competing AI labs may face stronger expectations to demonstrate evaluations, safeguards and accountability processes if they want to shape the summit's agenda credibly.
If expert warnings continue to be tied to deployment decisions, AI competition is likely to incorporate safety-testing capacity and governance readiness as strategic capabilities alongside model performance.
The pattern points toward recurring international coordination on frontier AI, though the corpus does not establish that summit discussions will produce binding shared rules.
The trend: Frontier-AI policy is shifting from broad warnings about misuse toward operational requirements for evaluating and governing dangerous capabilities.
https://www.theguardian.com/ ... And read the whole piece by the authors here .. https://managing-ai-risks.com/ There's always a danger of catastrophising new technology, but there are serious issues to understand here...
Together with 3 Turing Award winners (@geoffreyhinton, Yoshua Bengio, Andrew Yao), Nobel laureate @kahneman_daniel, Stuart Russel and 20 other AI & policy leaders, we just released a commentary that recommends actionable measures towards making AI safer: https://www.theguardian.c…
We note that, if managed carefully and distributed fairly, advanced AI systems could help humanity cure diseases, elevate living standards, and protect our ecosystems. But this piece necessarily focusses more on the many dangers of AI that urgently need to be addressed.
For big tech, the right type of regulation will provide economic incentives to shift resources from making the most capable systems yet more powerful to making them safer. But we need to avoid overregulation (e.g., in the EU AI Act) to help SMEs, startups and OSS initiatives.
23 AI experts, including @geoffreyhinton, Yoshua Bengio, @harari_yuval, @kahneman_daniel, @SheilaMcIlraith and Andy Yao, share a paper ahead of the UK's AI summit in Bletchley Park warning of worrying capabilities in current AI systems and seeking urgent governance:...
The short commentary draft is here: https://managing-ai-risks.com/ Coauthors include @dawnsongtweets @ancadianadragan @DavidSKrueger @pabbeel @jeffclune @ghadfield @harari_yuval @tegan_maharaj @SheilaMcIlraith & more. Led/organized by coauthors @JanMBrauner & @sorenmind