What happened
On 30 June 2026, Bank of England Deputy Governor Sarah Breeden delivered a speech at the ECB Sintra Forum ('Agents of Change') stating that human oversight is no longer a realistic primary defence against agentic-AI-driven systemic instability. She identified structural regulatory gaps — FCA's principles-based supervision cannot assign accountability when no human is in the decision loop; PRA operational resilience rules (e.g. PS21/3) address system downtime but not autonomous agent misbehaviour; and UK Market Abuse Regulation cannot address tacit algorithmic collusion between agents. She signalled forthcoming bespoke measures including mandatory kill-switch architecture, enhanced recovery capabilities, and possible market-wide circuit breakers to limit trading if faulty agentic models trigger disruption.
Why it matters
This is an authoritative signal from a G7 central bank that existing AI/operational-resilience frameworks (FCA principles, PRA PS21/3) are inadequate for agentic AI, foreshadowing binding rule-making for financial-sector agentic AI deployments — directly relevant to firms building autonomous trading, payments, or advisory agents.
Action needed
Financial institutions deploying agentic AI should begin scoping kill-switch/circuit-breaker architecture and accountability mapping ahead of anticipated BoE/PRA/FCA rule-making; monitor for a formal consultation.