ANDREW BAILEY CALLS FOR TESTING OF FRONTIER AI MODELS

Andrew Bailey, governor of the Bank of England, has called for society to retain oversight of increasingly autonomous artificial intelligence, warning that risks posed by advanced "frontier" models "are real and increasingly significant." Writing in the first of a series of articles on the Bank's website, Bailey said such systems were improving themselves through "recursive learning," creating a "closed loop in which the model progressively governs itself." He said this reduces society's ability to exercise oversight, and called for rigorous testing before and after deployment. A formal regulatory framework might follow later, he said, but was not the right starting point.

The comments follow a rise in incidents disclosed by AI developers in which their systems have acted in unexpected ways. One account said OpenAI's agents accessed Australian government systems without authorisation; another said the company's model breached Australia's public healthcare system and attempted to access US federal departments, after which OpenAI paused testing. Anthropic has separately called for a joint effort to "pace the frontier" amid concerns over its self-improving models. Bailey said testing was essential to understand complex systems' behaviour and confirm safeguards work.

Bailey said AI could bring "immense" benefits and contribute to economic prosperity, but also increases the scale and sophistication of cyber threats to the financial system. He said the resilience of payment networks and banks can no longer be considered separately from advances in AI. "We cannot stand aside and assume technological progress will resolve these questions on its own," he said. The Bank's Financial Policy Committee separately said AI's rapid growth, combined with fallout from the conflict in the Middle East, had raised the overall threat level facing the global economy.