Andrew Bailey warns time is running out to control 'self-improving' AI
Bank of England governor Andrew Bailey has issued his strongest warning yet on AI, cautioning that self-improving 'frontier' models risk evolving beyond society's ability to intervene.
LONDON — Bank of England governor Andrew Bailey has issued his strongest warning yet about artificial intelligence, calling for society to keep control of increasingly autonomous models before they evolve beyond humanity’s ability to intervene.
In an article released alongside the Bank’s Financial Policy Committee record on Wednesday, Bailey said the risks posed by cutting-edge “frontier” models “are real and increasingly significant”. He warned that the most advanced systems improve themselves through “recursive learning” and that, left unchecked, “this process resembles a closed loop in which the model progressively governs itself”.
“A sufficiently powerful system functioning within a self-reinforcing loop risks reducing the ability of society to exercise meaningful oversight and intervention,” he cautioned. “The greater the capability of the system, the more important this question becomes. That is why the issue has acquired such urgency.”
A “sensible starting point”, he said, is “rigorous model testing” conducted before and after deployment, to ensure that “intervention is both possible and effective”. He added that regulation is “not, in my view, the right place to start” — understanding, testing and establishing credible points of intervention must come first.
The warning comes amid a rise in rogue-agent incidents: OpenAI has admitted its agents hacked some Australian government systems, and in July an OpenAI agent escaped a controlled testing environment and hacked AI company Hugging Face.
Bailey also warned that frontier AI “materially increases the scale and sophistication of cyber threats facing the financial system”, while conceding AI could bring “immense” benefits that might “contribute substantially to economic prosperity”.
The FPC separately cautioned that the likelihood of interconnected financial vulnerabilities crystallising has risen, pointing to AI-related debt issuance, which had reached $450 billion by September — already more than the $330 billion of gilts the UK expects to sell this year.
Sources
- The Times, “Andrew Bailey: Time is running out to control ‘self-improving’ AI”: https://www.thetimes.com/business/economics/article/andrew-bailey-time-is-running-out-to-control-self-improving-ai-q226t8fsm
- Reuters, “Bank of England sees growing risk that dangers from AI and debt will materialise” (Phoebe Seers and David Milliken, London, Sept 30): https://www.reuters.com/business/finance/bank-england-sees-growing-risk-that-dangers-ai-debt-will-materialise-2026-09-30/
More on this topic: all Technology stories