Principals and Agents – speech by Jonathan Hall
SPEECH DIGEST
NEUTRAL high confidence · 34.3k characters readHall's valedictory FPC speech is a financial-stability essay on principal-agent risk as AI agents move into trading and market making, not a policy document. He explicitly disowns new regulation and frames the whole agenda as applying existing rules: algorithm identification under UK RTS 22, MiFID II disorderly-trading obligations, sandboxing, permissioning and chain-of-thought monitoring. There is no view on the policy path, no vote and no dissent, so nothing here moves the front end. The only market-pricing remark is a side note that a shortfall in AI capability would collide with the optimism embedded in asset prices, which he says is not his focus.
What’s new: Nothing material for rates. The new content is a scale reading and a regulatory posture: the BoE-FCA survey showing 80% of respondents using some form of agentic AI but only 2% with broad autonomy, and confirmation that autonomous trading by AI agents has begun at the retail end in crypto via MCP connections. Hall states plainly that he is not arguing for new regulations. The rest is his 2024 "Monsters in the Deep?" framework restated with more institutional detail and case studies.
KEY FINDINGS
- The Bank's joint survey with the FCA found 80% of respondents use some form of agentic AI, but only 2% of those agents have broad autonomy. The deployment curve is still early, so the systemic-behaviour risk is prospective rather than priced, which is why the speech carries no near-term market implication.
- Hall explicitly aligns with Governor Bailey in not seeking new regulation: the measures he details are designed to make AI agents comply with existing rules. No new compliance cost or regulatory shock to discount into bank, exchange or market-maker risk premia from this quarter.
- He flags the opposite tail: capabilities falling short of expectations would bring a different set of financial stability risks, particularly given the degree of optimism reflected in market prices. The only market-pricing signal in the document, and it points at equity AI-capex valuations, not at rates.
- Multi-agent dynamics are described as powerful and capable of harm: instrumental convergence, emergent hierarchy, tacit collusion and momentum ignition, the last group being hard to block through access permissions alone. A medium-term market-structure and supervision issue for venue and liquidity provision, not a policy-rate input.
- The proposed longer-term remedy is internalising the externality, using future models to price a market-efficiency impact that a verifier could block or an auditor could sanction. Hall labels this a research project, not an endorsed policy proposal. Signals the direction of FPC thinking without committing the Bank, so nothing to trade and nothing to hedge.
- Existing UK RTS 22 Article 8(3) unique algorithm designation is held up as already satisfying the AI-safety goal of a unique identity, enabling audit trail and revocation. The operational infrastructure for quarantining a rogue agent exists today, which lowers the tail risk of a disorderly AI-driven market event.
FROM THE DOCUMENT
The Bank’s recent survey with the FCA found that whilst 80% of respondents use some form of agentic AI only 2% of those agents have broad autonomy.
Consistent with this, and with recent remarks by Governor Andrew Bailey, I have not been arguing today for new regulations. Rather I have been detailing measures to ensure that the actions of AI agents comply with current regulations.
There is, of course, another possible future in which those capabilities fall significantly short of current expectations. That might bring a different set of financial stability risks, particularly given the degree of optimism reflected in market prices.
This remains, however, predominantly a discussion of future risks.
Given at CFM-Imperial Workshop on Quantitative Finance (hosted by Deutsche Bank)
In his final speech as an external member of the Financial Policy Committee, Jon Hall considers the growing use of AI agents in finance and the potential implications for financial stability. Drawing on principal-agent theory and the Bank of England’s own institutional framework, he discusses how access controls, monitoring and other safeguards could help manage the risks as AI agents become more capable. This will be my final speech as a member of the Bank of England’s Financial Policy Committee (“FPC”). Naturally, that makes me want to reflect on the lessons learned whilst working as a public servant at the Bank. But I also want to look forward and discuss one of the challenges the FPC is likely to face in the years ahead – the growing use of AI agents in finance.