BoE's Bailey: Test AI Before Regulating It — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Andrew Bailey, Governor of the Bank of England, wrote in his first-ever Substack article that regulating AI 'is not the right place to start,' calling instead for 'rigorous' testing to find vulnerabilities and build safeguards.
- Bailey acknowledged AI risks are 'real and increasingly significant' but said development should not be halted because 'the benefits are immense' — though a system for intervention and boundaries is still needed.
- Bailey singled out 'frontier AI' — highly-advanced, self-learning systems — warning that without effective oversight they risk becoming 'a closed loop in which the model progressively governs itself.'
- The UK's AI Security Institute is doing 'important work' in testing AI that should help form standards, per Bailey, though he said 'the pace of progress must accelerate.'
- Anthropic and OpenAI bosses have called for slowing AI development and an internationally co-ordinated risk approach; OpenAI recently declined to release its latest model over safety concerns.
- President Donald Trump rejected a slowdown, saying 'whoever wins AI, wins,' but later hosted executives from OpenAI, Anthropic, Nvidia, SpaceX, Meta and Google who signed a 'morally binding' document on AI risk.
- Bailey conceded testing will produce failures — models will 'behave unexpectedly' — but argued that is 'evidence of why testing is necessary,' and said regulation may still emerge over time.
Why it matters: Bailey's testing-first endorsement backs the UK's AI Security Institute as a model while creating a direct policy contrast with Trump's race-forward stance, leaving frontier AI's self-governing 'closed loop' risk to be caught by tests rather than rules. The burden falls on firms like OpenAI and Anthropic — whose own leaders are split between slowing development and declining to release models over safety concerns.
Ask SkimNews
