FCA Data Chief Jessica Rusu Outlines Agentic AI Supervision Plans

The Financial Conduct Authority said it will supervise autonomous AI in financial services under its existing rulebook, and asked AI providers to put clear disclaimers on financial content their models produce, in a speech by chief data, information and intelligence officer Jessica Rusu on October 19.

The speech, delivered at Bloomberg’s UK Policy Series, is the clearest statement yet of FCA agentic AI supervision. It announced no new AI-specific rules, no consultation and no timetable.

For compliance teams at banks and large firms, the message is that agents making or executing decisions fall under the same conduct, accountability and resilience expectations as any other system. The FCA also said it expects firms to report significant AI failures and near-misses.

What Rusu said on FCA agentic AI supervision

Rusu described AI moving toward increasingly autonomous systems. These agents search for products, manage payments and investments, and transact with other agents, according to the FCA’s published text of the speech. The FCA notes the text is a draft that may differ from what was delivered.

Greater autonomy raises the risk of consumer harm, cyber vulnerabilities and operational failures at scale, she said. Senior managers should be personally accountable for AI systems.

The FCA said it will not widen its remit. It intends to make the line between regulated and unregulated activity clearer.

“Our aim is to create the conditions in which agentic finance can thrive, with innovation and consumer protection built together from the start.”

Jessica Rusu, Chief Data, Information and Intelligence Officer, Financial Conduct Authority

Disclaimers on AI financial content

The FCA called on AI providers to add “clear, proportionate and consistent disclaimers” explaining the limits of AI-generated financial information and advice. The call is addressed to providers, many of which sit outside FCA authorization.

The regulator pointed to a gap in consumer understanding. Only 40% of consumers recognize that they have no route to redress if advice from a general-purpose AI tool goes wrong.

Use is already widespread among newer investors. Four in five less-experienced investors aged 18 to 40 have turned to AI for investing help, the FCA said.

Firms that build or distribute such tools remain responsible for what customers are told. A disclaimer from a model provider does not change that.

Accountability and assurance for autonomous AI

Rusu criticized firms that bolt oversight on after deployment or build tools without a defined purpose. The FCA’s companion publication, AI in financial services: good practice and areas for improvement, sets out examples gathered from firm testing this year.

It is organized around three themes: the Consumer Duty and consumers in vulnerable circumstances, governance and accountability, and operational resilience. The Consumer Duty is the FCA’s outcomes-based conduct standard for retail firms.

Examples of good practice include structured challenge, testing and sign-off before deployment, and stress-testing AI tools before they go live. Poor practice includes excessive reliance on human-in-the-loop controls, which can mask model weaknesses and leave accountability gaps.

It also lists overreliance on hard-to-replace AI providers as a resilience risk. Rusu praised one firm that built a live register of vulnerability signals so its chatbot could route at-risk customers to human advisers. She also cited another that set up an independent ethics panel to vet AI projects before launch.

Under FCA agentic AI supervision, agentic systems will need “continuous, conditional assurance,” she said, because a system that is safe at deployment may behave differently later. Existing incident reporting must extend to “significant AI failures and near-misses,” which she called a regulatory priority.

What is new and what is not

The FCA has said its agentic AI supervision does not depend on AI-specific regulation. It has relied on principles-based, outcomes-focused rules, chiefly the Consumer Duty and the Senior Managers and Certification Regime (SM&CR), which assigns named individuals responsibility for areas of a firm’s business.

Rusu’s speech did not name SM&CR, but its call for personal accountability for AI systems points to the same framework. The new elements are the guidance examples, the expectation on failure reporting, and the planned framework for agents.

That framework will be developed with industry, other regulators and frontier AI labs. It will cover agent identity, delegated authority, accountability and execution.

The speech also covered the Mills Review, published in July, which examined how AI could reshape retail financial services by 2030. The FCA is acting on six of its seven recommendations. It is not building the seventh, a public-interest AI-enabled financial capability service, and plans to build on existing work instead.

Rusu said the response is not to consult, study and report progress. It is to focus on testing, assurance, monitoring and surveillance. The good and poor practice examples will be revised as the technology develops, with no schedule given.

The FCA becomes an agentic regulator

Rusu said the FCA is reshaping itself as an agentic regulator. It is developing AI tools to supervise the 60,000 anti-money laundering (AML) firms that join its remit in 2028. It is also planning a market first responder agent for roughly 9,000 small wholesale transaction reporting firms.

The regulator launched its Handbook API in August, which lets firms ingest its rules in machine-readable form. It is now working on machine-readable data collection APIs.

For firms, that points to supervisory requests that machines can read and, over time, to data collected directly by system. Teams that already feed rules into agentic compliance tools will be able to point them at the FCA’s own text.

Adoption is already high. A joint FCA and Bank of England survey due the week after the speech finds that 86% of firms now use AI.

How the UK approach compares

The FCA’s route differs from the EU’s, where the EU AI Act sets binding obligations by risk category. UK firms operating in both markets will face one regime built on statute and another built on existing conduct and accountability rules.

Rusu said AI can increase productivity, improve competition and services, and drive economic growth if done well. “AI should improve consumer outcomes, not simply automate existing processes,” she said.

For AML teams, the 2028 expansion also changes the supervisor’s capacity. The role of the AML compliance officer will sit under a regulator that plans to use AI to oversee tens of thousands of additional firms.

What to watch next

The next fixed event is the publication of the joint FCA and Bank of England AI survey, due the week after the speech. No date has been given for the trusted framework for agentic finance or for the machine-readable data collection APIs.