Jonathan Hall, an external member of the Bank of England's Financial Policy Committee, warned Wednesday that handing market-making and other financial services to AI agents could concentrate risk among a few providers whose failures arrive together, according to the speech text the Bank published. The committee watches for risks to the stability of the U.K. financial system, so the speech gives banks, trading firms and asset managers an early read on how financial-stability officials are thinking about autonomous AI.

What Happened

Hall delivered "Principals and Agents" at the CFM-Imperial Workshop on Quantitative Finance, hosted by Deutsche Bank. He said it would be his final speech as a member of the committee.

He built the speech on the principal-agent problem, the economists' term for the risk that someone acting on another's behalf pursues different goals. In his framing, "the Bank and an AI agent are both delegates in a principal-agent relationship," so the Bank's own arrangements for staying transparent, accountable and constrained offer a template for controlling agents. Delegating to powerful agents is not new, he said, but the speed and scale at which AI agents act and interact mean some guardrails will need to evolve. He called for managing risk through both top-down supervision and bottom-up controls built into the models.

RelatedNonprofits Sue OpenAI, Anthropic, Google, Meta and SpaceXSI to Halt New AI Models

At the level of a single firm, Hall said harm can be curbed through access restrictions, diligent monitoring and a unique identity for each agent. He said reading a model's chain of thought, the step-by-step reasoning some AI systems write out, has helped both performance and safety but is "somewhat fragile." Watching agents' actions and attempted actions remains essential, and firms will need oversight that is scalable and risk-based to avoid bottlenecks and alert fatigue.

Why It Matters

Hall's main concern is the system as a whole. Using liquidity provision in systemically important markets such as government bonds as a case study, he named three dynamics: more concentration among service providers, more correlation between them overall and more correlation in times of stress. The lesson of the financial crisis, he said, is that failures cluster in downturns. On whether delegating such services to AI agents will intensify those dynamics, he said, "we don't know, but there is a risk that it will."

He also cautioned that even if every firm gets its controls right, that does not guarantee a stable financial system. He noted, without making it his focus, that AI falling short of today's market optimism would carry its own stability risks.

The controls Hall described echo international guidance. The Primary Wire reported in October that the Financial Stability Board told banks to give each AI agent a unique identifier, limited access and a tested kill switch. For U.S. boards, the speech adds to signals that supervisors on both sides of the Atlantic are asking the same questions; The Primary Wire reported, citing Reuters, that U.S. banking regulators have asked banks how they monitor, audit and shut down autonomous AI. Firms that trade in London, including U.S. banks, would feel any shift in British supervisory expectations directly.

Context

Hall's speech follows remarks by Nikhil Rathi, chief executive of the Financial Conduct Authority, Britain's markets regulator, who said in October that AI agents will test "when agentic coordination becomes market manipulation," as The Primary Wire reported.

What to Watch

Hall urged more experimental and theoretical analysis. He pointed to Project Logos, which the Bank for International Settlements has just launched to compare simulated portfolio managers run on fixed rules with ones run by large language models, tracking how their decisions diverge and correlate in good and bad times. He also suggested tracking market concentration over time with standard measures such as the Herfindahl-Hirschman Index. The next test is whether the Financial Policy Committee or Britain's regulators turn those ideas into formal expectations for firms.