# FCA and AI Agents: The Rules That Already Apply > Source: > Published: 2026-10-27 00:00:00+00:00 [← All posts](https://reinvently.co.uk/blog/) ## Key Takeaways - On 19 October 2026 the FCA promised "a trusted framework for agentic finance". It has published no draft rules, consultation or timetable. - Agents are already live. In the Bank of England and FCA survey of 26 October, 80 percent of respondents had deployed AI agents. Only 2 percent gave them broad autonomy. - Four regimes you already follow cover what an agent does: SM&CR, outsourcing, operational resilience and, for retail customers, the Consumer Duty. - The gaps are about whose agent it is and what it may do: agent identity, delegated authority and assurance after launch. - New incident and third-party reporting rules apply from 18 March 2027. The FCA wants incident reporting to "extend to significant AI failures and near-misses". **Sources, checked 27 October 2026:** Jessica Rusu's speech [Building the future of finance in the age of AI](https://www.fca.org.uk/news/speeches/building-future-finance-age-ai) (19 October 2026); the FCA's [AI in financial services: good practice and areas for improvement](https://www.fca.org.uk/publications/good-poor-practice/ai-financial-services) (19 October 2026); [The Mills Review](https://www.fca.org.uk/publications/corporate-documents/mills-review) (6 July 2026); the Bank of England and FCA report [Artificial intelligence in UK financial services – 2026](https://www.bankofengland.co.uk/report/2026/artificial-intelligence-in-uk-financial-services-2026) (26 October 2026); [PS26/2: Operational incident and third party reporting](https://www.fca.org.uk/publications/policy-statements/ps26-2-operational-incident-third-party-reporting) (18 March 2026); and [FinTech Global's report](https://fintech.global/2026/10/26/fca-builds-agentic-finance-rules-as-ai-reshapes-uk-markets/) (26 October 2026). On 26 October, FinTech Global ran the headline "FCA builds agentic finance rules as AI reshapes UK markets". The FCA's own documents say something narrower. There is no draft rulebook, no consultation and no date. What exists is a promise. On 19 October Jessica Rusu, the FCA's Chief Data, Information and Intelligence Officer, said the regulator would build "a trusted framework for agentic finance". It will do so with industry, other regulators and frontier AI labs. Your agents will not wait for it. In the Bank of England and FCA survey published on 26 October, 80 percent of respondents had deployed AI agents. Only 2 percent gave them broad autonomy, so most agents still support decisions rather than take them. The FCA's Mills Review expects that to change: AI "will recommend actions, initiate transactions, and execute decisions within agreed parameters". Until the framework arrives, the rules you already follow decide who answers for an agent. The FCA says so directly: firms using AI tools "remain responsible for consumer outcomes". ## What the FCA Has Said, and What It Has Not - **A framework, with no date.** At Bloomberg's UK Policy Series on 19 October, Rusu promised "clear approaches to agent identity, delegated authority, accountability and execution". She gave no timetable and announced no consultation. - **Good practice, not new rules.** The FCA's good and poor practice examples, published the same day, "do not create new requirements". Two concern agents: firms set up governance before deploying more autonomous agents, and kept a central inventory of them. - **Six of seven Mills Review recommendations.** The FCA is acting on six, including "Enable the foundations for agentic finance". It declined the seventh, a public-interest AI service for financial capability. - **A clearer boundary, not a wider remit.** "Our answer is not simply to expand our remit, but to make the existing boundary clearer," Rusu said. How the FCA will treat consumer agents run by unregulated companies is still open. - **Incident reporting, in principle.** Rusu said reporting of major incidents "must extend to significant AI failures and near-misses". She did not say whether that needs a rule change. The framework is a statement of intent. The obligations are already in force. ## Four Regimes That Already Cover an Agent's Actions An agent that moves money, settles a claim or refuses an application is doing regulated work. Four regimes already apply to that work, whoever or whatever does it. Each asks a different question. | Regime | What it already requires | The question an agent raises | |---|---|---| | Senior Managers and Certification Regime | A named senior manager answers for each business area and must take reasonable steps to prevent breaches | Which senior manager owns this agent, and what steps can they show? | | Outsourcing and third parties (SYSC 8, PS26/2) | The firm stays responsible for what it outsources; from 18 March 2027, material third-party arrangements are reported to the FCA | Is the model or agent provider a material third party, and will you hear when its model changes? | | Operational resilience (PS21/3) | Important business services stay within impact tolerances in severe but plausible scenarios, required since 31 March 2025 | Does an important business service now depend on an agent or its model? | | Consumer Duty (PRIN 2A) | Good outcomes for retail customers on products, price and value, understanding and support | Have the agent's outcomes been tested across customer groups, including vulnerable customers? | ### Senior Managers and Certification Regime Under the duty of responsibility, a senior manager can be held accountable for a breach in their area. The test is whether they took reasonable steps to prevent it. An agent does not change whose area it is. The FCA's good practice examples describe senior managers who "owned risks from the design stage onwards for each AI use case". It does not expect them to be technical specialists. In practice, an agent that moves customers to better savings rates sits with whoever runs savings. That person needs evidence of reasonable steps: the agent's mandate, its limits, its test results and who reviews its exceptions. Human sign-off alone may not count. The FCA warns that human review can become "a 'box-ticking' exercise, rather than meaningful oversight". ### Outsourcing and third parties The outsourcing rules in SYSC 8 let you hand over a function but not the responsibility for it. The FCA applies that to AI in one line: "Firms can outsource AI services, but remain responsible for them." Agents deepen the dependency. In the survey, third parties supplied 53 percent of high-materiality AI systems, up from 35 percent in 2024. Two things follow. First, a model update can change what your agent does without any change to your code. The FCA's good practice examples include contracts that require notice of significant model changes. Second, from 18 March 2027 PS26/2 makes you report material third-party arrangements to the FCA. A provider whose model now executes payments may belong on that register. ### Operational resilience Since 31 March 2025, firms must keep each important business service within its impact tolerance in severe but plausible scenarios. The FCA lists "failure to consider AI when reviewing operational resilience" as an area for improvement. Once an agent handles part of a claims or payments service, it belongs in that service's map. The failure modes are new. A provider outage is familiar. A model update that changes behaviour, or an agent stuck in a loop, is not. As Rusu put it, a system "safe at deployment may behave differently at scale, under stress or as the models beneath it change". Test those scenarios, and keep a fallback that holds the service within tolerance. ### Consumer Duty The Consumer Duty has applied to open products since 31 July 2023. It sets four outcomes for retail customers: products and services, price and value, understanding and support. An agent is one more channel, and the outcomes remain yours. The FCA's areas for improvement show where agents will struggle: "Assuming AI outputs are clear without testing them" and "Excluding consumers in vulnerable circumstances by default". An agent that switches a product must deliver fair value and an explanation the customer understands. Customers who need a person still need a route to one. The Duty covers retail customers only, so wholesale agents rely on the other three regimes. ## Where the Existing Rules Run Out The four regimes assume your own systems act for you. Agentic finance breaks that in three places, and they match Rusu's agenda. - **Agent identity.** A customer's own agent, run by a technology company, may instruct your firm. You need to know whose agent it is and whether the instruction is genuine. None of the four regimes says how to check. - **Delegated authority.** A customer might tell an agent to move their savings whenever a better rate appears. No rule says how that mandate is recorded, limited or revoked, or who bears the loss if the agent exceeds it. - **Assurance after launch.** Rusu called for "continuous, conditional assurance". That is where firms are weakest. In the survey, more than 90 percent of respondents reported human approval and pre-deployment review, but only around half reported ongoing monitoring. Waiting for the framework is tempting. But the duties above apply today, and any enforcement case will be judged against them. Firms running agents now also hold the evidence the FCA needs to shape what comes next. ## What to Do Before the Framework Arrives 1. **Register every agent and name its owner.** Keep a central inventory, as the FCA's examples describe. Record the senior manager accountable for each agent, its purpose, and the data and systems it can reach. 2. **Enforce each agent's mandate outside the model.** Set transaction limits, products and counterparties in tool permissions and payment controls, not in the prompt. That record is your evidence of delegated authority. 3. **Map agents into operational resilience.** Note where an agent or model sits in each important business service. Test what happens when it fails or changes behaviour. 4. **Fix contracts and the third-party register.** Ask AI providers for notice of significant model changes. Add material AI providers to the register you will report from 18 March 2027. 5. **Test outcomes and log near-misses.** Test agent outcomes across customer groups, including vulnerable customers, and keep a human route. Start recording AI near-misses now. For a fuller pre-production checklist, see the [FSB's 12 AI practices](https://reinvently.co.uk/blog/fsb-ai-practices-uk-financial-firms/). Our [agent incident log](https://reinvently.co.uk/tools/agent-incidents/) shows the kind of failures near-miss reporting will need to capture. ## Frequently asked questions ### Has the FCA published rules for AI agents? No. On 19 October 2026 it promised a framework for agent identity, delegated authority, accountability and execution, to be built with industry, other regulators and AI labs. It has published no draft rules, consultation or timetable. Its good practice examples on AI create no new requirements. ### Does the Consumer Duty apply when an AI agent makes the decision? Yes, for retail customers. The FCA says firms using AI tools remain responsible for consumer outcomes. An agent has to deliver the same outcomes on products, price and value, understanding and support as any other channel. ### Who is accountable under SM&CR when an AI agent makes a mistake? The senior manager responsible for the area where the agent acts. They can be held accountable for a breach if they did not take reasonable steps to prevent it. The FCA does not expect them to be technical specialists. ### When do the FCA's new incident and third-party reporting rules apply? From 18 March 2027, under PS26/2. Firms must report operational incidents above set thresholds and report their material third-party arrangements to the FCA. In October 2026 the FCA said incident reporting must extend to significant AI failures and near-misses. *We are following the FCA's agentic finance work alongside the ICO's. [Get new posts by email](https://reinvently.co.uk/contact/).* [← All posts](https://reinvently.co.uk/blog/)