AI/News
Bank of England governor: society must keep the power to intervene in frontier AI - test first, regulate later
In a Bank Insights post Andrew Bailey argues recursive learning risks a closed loop 'in which the model progressively governs itself', and names cyber threats to the financial system as the immediate concern. No policy is announced.
By Daily Aletheia · Checked against the primary source · 30 September 2026 · 2 min read
Andrew Bailey, governor of the Bank of England, published "Frontier AI and the question of governance" on the Bank's Insights blog on 30 September.
The argument
Bailey writes that the risks from frontier AI "are real and increasingly significant", but that regulation is not the first question. The first question is "what precisely is the problem we are trying to solve?"
His answer turns on what he calls recursive learning. Frontier systems "increasingly use their own outputs to refine and improve their performance", a feedback process that can become self-referential. "Without effective means of intervention, this process resembles a closed loop in which the model progressively governs itself."
A sufficiently powerful system in that loop, he says, "risks reducing society's ability to exercise meaningful oversight and intervention". Should society retain the ability to intervene and set the boundaries? "To my mind, the answer is unequivocally yes."
What he proposes
Rigorous model testing before and after deployment, built into standards for deploying models, and learning from incidents including near misses. He credits the UK's AI Security Institute with "a strong start" but says "the pace of progress must accelerate".
Testing is not offered as a substitute for rules: "Over time, a more formal regulatory framework may well emerge. But regulation is not, in my view, the right place to start."
Why a central bank
Bailey says frontier AI "materially increases the scale and sophistication of cyber threats facing the financial system", and that the resilience of payments networks, market infrastructure and banks can no longer be considered separately from AI. Testing the models firms use could, in time, be codified into standards across the financial system.
The post does not announce any Bank policy, requirement or timetable.
The Bank published an explainer on AI, jobs and wages the same day. Its one hard figure: nearly 90% of firms in the Bank's own survey said AI had made no noticeable difference to the number of people they employ.