What Sarah Breeden Actually Said
On June 30, 2026, Bank of England Deputy Governor Sarah Breeden addressed the ECB Forum on Central Banking in Sintra, Portugal. Her remarks on agentic AI were not a cautionary note at the margins of a broader speech. They were the central argument. Breeden called explicitly for circuit breakers and kill switches on AI agents operating in financial markets, and she framed the governance challenge in terms that every enterprise board should take seriously.
Her core argument was this: AI task-completion capability is doubling every four months. Fifty-two percent of financial firms are already using agentic AI. And the regulatory frameworks designed to govern human decision-making in financial markets were not built to handle systems that can take thousands of autonomous actions per second, interact with other AI systems across institutional boundaries, and pursue objectives over extended time periods without continuous human supervision.
The Bank of England's position is not that agentic AI should be restricted. It is that agentic AI operating without adequate governance infrastructure creates systemic risk that no single institution's internal controls can manage. The circuit breaker and kill switch framework is the regulator's response to that gap.
Why Agentic AI Is a Different Risk Category
Most enterprise risk frameworks treat AI as a variant of existing technology risk: software that can produce incorrect outputs, create data privacy issues, or introduce operational dependencies on third-party vendors. These risks are real, but they are manageable within existing governance structures. Agentic AI introduces a qualitatively different risk category that existing frameworks were not designed to address.
The difference is autonomy and interconnection. An agentic AI system does not wait for human instruction before taking the next action. It pursues an objective through a sequence of steps, interacting with other systems, making decisions, and adapting its approach based on the results it receives. In a financial context, this means an AI agent can execute a complex trading strategy, interact with counterparty AI systems, and create significant market positions before any human has reviewed its actions.
The systemic risk dimension that concerns the Bank of England is what happens when multiple AI agents from different institutions behave in correlated ways during market stress. If several major banks deploy AI agents trained on similar data with similar objectives, those agents may respond to the same market signals in the same way simultaneously, amplifying volatility rather than absorbing it. This is the circuit breaker problem: no single institution's risk management detects the systemic pattern because each institution only sees its own agent's behaviour.
The Three Governance Requirements
Breeden's framework identifies three specific governance requirements for agentic AI in financial markets. The first is automated circuit breakers: mechanisms that pause or halt AI agent activity when predefined risk thresholds are breached. These are not manual override buttons. They are automated controls embedded in the AI system's operating parameters that trigger without human intervention when the system's behaviour exceeds authorised boundaries.
The second requirement is human-in-the-loop for high-stakes decisions. This does not mean human approval for every action an AI agent takes. It means identifying the categories of decision where AI agent errors would have material financial, regulatory, or reputational consequences, and requiring human review before those decisions are executed. The governance challenge is defining those categories with sufficient precision that the human-in-the-loop requirement is meaningful rather than performative.
The third requirement is systemic risk monitoring: the infrastructure to detect when multiple AI agents are behaving in correlated ways that could amplify market instability. This is the most technically demanding requirement because it requires visibility across institutional boundaries. The Bank of England's implicit message is that regulators will need to develop the monitoring infrastructure themselves if institutions do not build it collaboratively.
What This Means for Enterprise Boards
The Bank of England's governance framework is directed at financial institutions, but its implications extend to every enterprise board deploying agentic AI. Financial services regulators consistently set the governance standards that other regulators follow, and the circuit breaker framework will become the template for agentic AI governance requirements across sectors.
For enterprise boards, the immediate implication is that agentic AI governance is now a fiduciary matter, not an IT matter. The question of whether your organisation has adequate circuit breakers, kill switches, and human-in-the-loop requirements for its AI agents is a question that directors can be held accountable for. The organisations that build this infrastructure proactively will be better positioned when sector-specific regulation arrives than those that wait for formal requirements.
For agentic AI deployment, this means that the governance infrastructure is not a constraint on AI capability. It is the condition that makes expanded AI autonomy commercially and regulatorily sustainable. The organisations that demonstrate robust governance to regulators are the ones that will be permitted to deploy more capable AI systems with less supervisory friction. The governance investment is the price of the competitive advantage.
The DORA AI compliance framework provides a practical starting point for financial services organisations building this governance infrastructure. For enterprises outside financial services, the Claude and Mythos cybersecurity analysis illustrates the attack surface that inadequate AI governance creates. The Bank of England's circuit breaker framework is the regulatory response to both dimensions of that risk.
If your board's risk committee does not yet have agentic AI on its agenda as a distinct governance item, the Bank of England's June 2026 speech is the document to circulate. The question is not whether agentic AI governance will become a regulatory requirement. It is whether your organisation will have the infrastructure in place before the requirement arrives. The free AI growth audit maps your current agentic AI deployment against the emerging governance standard and identifies the gaps that represent the highest regulatory and commercial risk.
About the Author
Modi Elnadi is Founder and Director of Marketing and AI Growth at Integrated.Social, a London-based B2B AI growth marketing agency. Modi works with enterprise and financial services clients on agentic AI governance frameworks, helping organisations build the oversight infrastructure that satisfies regulatory expectations while preserving commercial velocity. His work sits at the intersection of AI deployment, risk governance, and the commercial decisions that determine whether AI transformation creates durable competitive advantage or unmanaged regulatory exposure. He writes regularly on AI governance, financial regulation, and the enterprise implications of central bank and regulatory policy on AI adoption.
What Sarah Breeden Actually Said
On June 30, 2026, Bank of England Deputy Governor Sarah Breeden addressed the ECB Forum on Central Banking in Sintra, Portugal. Her remarks on agentic AI were not a cautionary note at the margins of a broader speech. They were the central argument. Breeden called explicitly for circuit breakers and kill switches on AI agents operating in financial markets, and she framed the governance challenge in terms that every enterprise board should take seriously.
Her core argument was this: AI task-completion capability is doubling every four months. Fifty-two percent of financial firms are already using agentic AI. And the regulatory frameworks designed to govern human decision-making in financial markets were not built to handle systems that can take thousands of autonomous actions per second, interact with other AI systems across institutional boundaries, and pursue objectives over extended time periods without continuous human supervision.
The Bank of England's position is not that agentic AI should be restricted. It is that agentic AI operating without adequate governance infrastructure creates systemic risk that no single institution's internal controls can manage. The circuit breaker and kill switch framework is the regulator's response to that gap.
Why Agentic AI Is a Different Risk Category
Most enterprise risk frameworks treat AI as a variant of existing technology risk: software that can produce incorrect outputs, create data privacy issues, or introduce operational dependencies on third-party vendors. These risks are real, but they are manageable within existing governance structures. Agentic AI introduces a qualitatively different risk category that existing frameworks were not designed to address.
The difference is autonomy and interconnection. An agentic AI system does not wait for human instruction before taking the next action. It pursues an objective through a sequence of steps, interacting with other systems, making decisions, and adapting its approach based on the results it receives. In a financial context, this means an AI agent can execute a complex trading strategy, interact with counterparty AI systems, and create significant market positions before any human has reviewed its actions.
The systemic risk dimension that concerns the Bank of England is what happens when multiple AI agents from different institutions behave in correlated ways during market stress. If several major banks deploy AI agents trained on similar data with similar objectives, those agents may respond to the same market signals in the same way simultaneously, amplifying volatility rather than absorbing it. This is the circuit breaker problem: no single institution's risk management detects the systemic pattern because each institution only sees its own agent's behaviour.
The Three Governance Requirements
Breeden's framework identifies three specific governance requirements for agentic AI in financial markets. The first is automated circuit breakers: mechanisms that pause or halt AI agent activity when predefined risk thresholds are breached. These are not manual override buttons. They are automated controls embedded in the AI system's operating parameters that trigger without human intervention when the system's behaviour exceeds authorised boundaries.
The second requirement is human-in-the-loop for high-stakes decisions. This does not mean human approval for every action an AI agent takes. It means identifying the categories of decision where AI agent errors would have material financial, regulatory, or reputational consequences, and requiring human review before those decisions are executed. The governance challenge is defining those categories with sufficient precision that the human-in-the-loop requirement is meaningful rather than performative.
The third requirement is systemic risk monitoring: the infrastructure to detect when multiple AI agents are behaving in correlated ways that could amplify market instability. This is the most technically demanding requirement because it requires visibility across institutional boundaries. The Bank of England's implicit message is that regulators will need to develop the monitoring infrastructure themselves if institutions do not build it collaboratively.
What This Means for Enterprise Boards
The Bank of England's governance framework is directed at financial institutions, but its implications extend to every enterprise board deploying agentic AI. Financial services regulators consistently set the governance standards that other regulators follow, and the circuit breaker framework will become the template for agentic AI governance requirements across sectors.
For enterprise boards, the immediate implication is that agentic AI governance is now a fiduciary matter, not an IT matter. The question of whether your organisation has adequate circuit breakers, kill switches, and human-in-the-loop requirements for its AI agents is a question that directors can be held accountable for. The organisations that build this infrastructure proactively will be better positioned when sector-specific regulation arrives than those that wait for formal requirements.
For agentic AI deployment, this means that the governance infrastructure is not a constraint on AI capability. It is the condition that makes expanded AI autonomy commercially and regulatorily sustainable. The organisations that demonstrate robust governance to regulators are the ones that will be permitted to deploy more capable AI systems with less supervisory friction. The governance investment is the price of the competitive advantage.
The DORA AI compliance framework provides a practical starting point for financial services organisations building this governance infrastructure. For enterprises outside financial services, the Claude and Mythos cybersecurity analysis illustrates the attack surface that inadequate AI governance creates. The Bank of England's circuit breaker framework is the regulatory response to both dimensions of that risk.
If your board's risk committee does not yet have agentic AI on its agenda as a distinct governance item, the Bank of England's June 2026 speech is the document to circulate. The question is not whether agentic AI governance will become a regulatory requirement. It is whether your organisation will have the infrastructure in place before the requirement arrives. The free AI growth audit maps your current agentic AI deployment against the emerging governance standard and identifies the gaps that represent the highest regulatory and commercial risk.
About the Author
Modi Elnadi is Founder and Director of Marketing and AI Growth at Integrated.Social, a London-based B2B AI growth marketing agency. Modi works with enterprise and financial services clients on agentic AI governance frameworks, helping organisations build the oversight infrastructure that satisfies regulatory expectations while preserving commercial velocity. His work sits at the intersection of AI deployment, risk governance, and the commercial decisions that determine whether AI transformation creates durable competitive advantage or unmanaged regulatory exposure. He writes regularly on AI governance, financial regulation, and the enterprise implications of central bank and regulatory policy on AI adoption.




