Bank of England: Current Regulation Not Designed for Agentic AI in Finance

The Bank of England in July 2026 announced the beginning of an official review of regulatory frameworks to determine whether existing rules cover the use of agentic AI in the financial sector — in payment systems, securities trading, cybersecurity, and operational activities.

Why existing rules are not designed for agentic AI

Sarah Breeden, deputy governor of the Bank of England, addressed the European Central Bank's central banking forum with a direct acknowledgment: existing regulatory frameworks were created without accounting for AI systems capable of making decisions and taking action without direct human participation at each step.

Agent systems represent a fundamentally different class of software. Unlike traditional algorithms or early language models, agents can independently decompose tasks, call external tools, and initiate real actions: execute payments, conduct trading transactions, adjust risk parameters, or respond to cyberattacks — without explicit operator confirmation at each step. It is this autonomy that creates a gap between the letter of existing regulations and the practice of AI application.

• Review scopes: payments, trading, cybersecurity, operational processes
• Breeden's key thesis: existing frameworks were not designed for autonomous AI agents
• Venue: ECB central banking forum, July 2026

Where exactly does regulatory gap arise

Most modern financial regulations are built on a fundamental principle: each significant financial decision must have an identifiable person responsible for it and capable of explaining it. Regulators require documenting who made the decision, on what basis, with what authorities, and within what supervisory framework.

Agent systems complicate this logic to the extreme. A single transaction or trading operation may include dozens of autonomous steps performed by several interacting agents: one collects data, another forms a hypothesis, a third executes the deal, a fourth manages risk. Where in this chain is the "responsible person"? From the perspective of existing legislation — nowhere.

Additional complexity comes from speed. Agent systems in trading and cybersecurity make decisions in milliseconds. Full real-time human oversight is physically impossible — yet this is what most existing liability rules assume.

"Existing regulatory frameworks were not designed for AI agents capable of acting without direct human participation," —

Sarah Breeden, deputy governor of the Bank of England.

What this means

The Bank of England is one of the world's most influential financial regulators, setting standards that other jurisdictions look to. Breeden's public acknowledgment at the ECB forum is not an academic observation, but an official signal: the regulator sees the problem and is starting to act.

For financial companies already deploying AI agents in trading systems, payment infrastructure, and security services, this means: the period of quiet regulatory uncertainty is ending. The outcome — either targeted clarifications of existing rules, or fundamentally new requirements for autonomous systems. It's better to prepare now.

*Meta is recognized as an extremist organization and prohibited in the Russian Federation.