The Bank of England is preparing strict rules for AI agents in finance: cyberattacks, herd behavior, and payment failures in the crosshairs.
The Bank of England has officially announced a review of the regulatory framework for artificial intelligence in the financial sector. Deputy Governor Sarah Breeden, speaking at the European Central Bank forum in Sintra, clearly stated that current technology-neutral rules are not designed for systems capable of functioning without human intervention. The main risks, according to her, are cyberattacks, failures in payment infrastructure, and synchronous market errors.
A Dual Problem: Implementation and Transformation
Breeden emphasized that the acceleration of AI development, especially over the past six months, creates a dual challenge. On one hand, it is necessary to ensure responsible implementation and risk management. On the other, it must be acknowledged that AI should fundamentally change the working methods of central banks themselves. "We need to seriously consider both the first and the second aspects," she stated.
Specific Measures: Isolation, Recovery, and "Emergency Stop"
Among the potential tools, the regulator is considering tightening requirements for the recovery of key systems, creating fully isolated backup capacities, and the ability to quickly restructure compromised infrastructure. For markets, the introduction of safeguards similar to trading circuit breakers is being explored — in case erroneous AI models trigger large-scale disruptions. Breeden specifically highlighted the risk of herd behavior, where agents react to identical triggers, amplifying volatility.
Context and Global Coordination
This statement is part of a broader trend. On May 15, the Bank of England, the Financial Conduct Authority, and HM Treasury issued a joint statement on the cyber risks of advanced AI models. On June 22, the Five Eyes intelligence agencies gave a similar assessment, warning that the window between vulnerability discovery and exploitation is shrinking to months. On June 10, the Financial Stability Board published a consultation paper with 12 practices for responsible AI implementation, and in May 2026, the IMF called for considering AI cyber risks as a matter of financial stability.
Adoption Statistics: 81% Already in Play
According to data from the Cambridge Centre for Alternative Finance, 81% of industry participants have already adopted AI at some level, and 52% are in the pilot or more mature deployment stage of agentic AI. So far, the main application is in internal processes: automation, data visualization, and software development. Adoption in the front office remains limited, but as practice shows, it is only a matter of time.
In a response to a parliamentary committee on April 1, the Bank of England acknowledged that generative and agentic AI do not yet pose a systemic risk but emphasized that the technology-neutral approach will be reviewed. The discussion is also ongoing in other institutions, including the Bundesbank and the Bank for International Settlements.
My analysis: The Bank of England is acting proactively, which is rare for regulators. The problem is not that AI agents are already breaking markets, but that their mass adoption is inevitable, and "technology-neutral" rules are a fiction in a world where algorithms make decisions faster than humans. If the regulator actually introduces isolated backup capacities and "emergency stops," this will set a precedent for the entire global financial sector. The only question is whether they will manage to do so before the first major failure occurs on its own.