AI in action · Regulation

BoE's Jon Hall: Treat AI Agents Like Any Powerful Delegate

In his final speech as an FPC member on November 11, 2026, Jon Hall said AI agents in finance should obey the same rules as human traders and be sandboxed. He noted that only 2% of agents in a BoE-FCA survey had broad autonomy.

Jon Hall used his last speech as an external member of the Bank of England's Financial Policy Committee to look at a problem the committee expects to face for years: AI agents that act on behalf of investors and firms. Speaking on November 11, 2026 at the CFM-Imperial Workshop on Quantitative Finance, hosted by Deutsche Bank, he framed it as an old question, how to delegate power safely, and used the Bank itself as the example. "The Bank and an AI agent are both delegates in a principal-agent relationship," he said.

He was clear that this is mostly about the future. Since his 2024 speech "Monsters in the Deep?", he said, autonomous trading by AI agents has begun, "led by retail investors in non-core markets like crypto" and enabled by MCP connections. But the Bank's recent survey with the FCA found that while 80% of respondents use some form of agentic AI, only 2% of those agents have broad autonomy.

His prescriptions are concrete. AI agents should be trained "to respect the very same rules and regulations as a human employee." For high-stakes uses they should run inside a sandbox that limits what they can do, configured on the principle of least privilege, though he noted sandboxes can be broken. To trade, an agent should need credentials tied to a unique identity, building on the UK's existing requirement that every trading algorithm carries its own identifier. And both an agent's internal reasoning and its external actions should be monitored where possible.

The financial-stability worry is correlation. Hall asked whether handing financial services to AI agents would increase concentration among providers or make firms behave alike, warning that common model failures "could cause them to all have to reduce the same risk at the same time (as in a flash crash or the bursting of a credit bubble)." He added that the collective behaviour of AI agents is "very poorly understood" and pointed to MiFID II's rule that algorithms must not contribute to disorderly trading and must be able to be switched off in stress.

He also flagged the opposite risk without dwelling on it: that AI capabilities fall well short of expectations, which "might bring a different set of financial stability risks, particularly given the degree of optimism reflected in market prices." For traders already handing tasks to agents, his checklist is a useful one: clear limits, a log of every action, a way to stop the agent and someone accountable for what it does.

Sources: Bank of England — Principals and Agents, speech by Jonathan Hall (Nov 11, 2026)

By Trade2Spike Team, in-house editorial team. About us · Methodology