---
title: What does the UK FCA expect from firms using AI agents?
url: https://opencontroller.lyzr.com/blog/uk-fca-expectations-for-ai-agents
published: 2026-11-16
updated: 2026-11-16
author: Lyzr
---

# What does the UK FCA expect from firms using AI agents?

**The FCA has no AI-specific rulebook and has said it does not plan to add one. It expects firms to meet existing rules when they use AI agents: the Consumer Duty for customer outcomes, the Senior Managers and Certification Regime (SM&CR) for named accountability, and existing systems-and-controls and operational resilience rules. The July 2026 Mills Review asked the FCA to clarify how those rules apply to more autonomous AI, and to lead a trusted framework for AI agents.**

For a UK-regulated firm, the question is not "which AI law applies?" but "how do we show the rules we already follow still hold when an agent is doing the work?" This guide summarises the FCA's current position, the Mills Review's recommendations on agents, and the evidence a firm can assemble. It is a summary, not legal advice. For the governance model underneath, see [what is an AI agent control plane](/blog/what-is-an-ai-agent-control-plane).

## What is the FCA's current position on AI?

The FCA takes a principles-based, outcomes-focused approach. On its Mills Review page it states it does not plan to introduce extra regulations for AI and will rely on existing frameworks. In practice that means:

- **Consumer Duty** (PRIN 2A): firms must deliver good outcomes for retail customers, including on products and services, price and value, consumer understanding and consumer support.
- **SM&CR**: a named senior manager is accountable for each business area, and must take reasonable steps to prevent breaches in it. AI used in that area falls within their responsibility.
- **Systems and controls** (SYSC), including outsourcing and third-party risk.
- **Operational resilience**: important business services must stay within impact tolerances, including where they rely on AI services.

Banks and insurers dual-regulated by the PRA should also read the PRA's model risk management principles (SS1/23), which took effect in May 2024.

## What did the Mills Review say about AI agents?

The FCA Board commissioned Sheldon Mills to review how AI could reshape retail financial services by 2030. The timeline:

| Date | Step |
|---|---|
| 27 January 2026 | Engagement paper published as a call for input |
| 24 February 2026 | Call for input closed |
| 6 July 2026 | Mills Review published |

The Review described a shift where staff move from operators to approvers and observers of AI. Two recommendations matter most for agents:

1. **Clarify existing rules.** The FCA should clarify how the Consumer Duty and SM&CR apply as AI becomes more autonomous. Respondents did not ask for AI-specific rule changes.
2. **A trusted framework for AI agents.** The FCA should lead a framework for AI agent participation in financial services covering identity, authority, accountability and execution, with trusted agent standards suggested through Open Finance.

The Review proposed no rule changes. Its analysis is expected to inform supervisory priorities and future guidance, so watch for FCA publications that follow it.

## What is AI Live Testing?

AI Live Testing is an FCA initiative where selected firms test AI applications in live conditions with FCA support, focusing on risk management and live monitoring. The FCA announced a second cohort of eight firms on 21 April 2026, after applications opened in January 2026. Agentic AI appears among the model types in applications, and agentic payments among the use cases. It is a supervisory engagement, not a certification.

## What evidence should a firm keep for each AI agent?

Map each FCA expectation to evidence you can produce on request:

| FCA expectation | Question a supervisor may ask | Evidence for it |
|---|---|---|
| Consumer Duty outcomes | Does the agent produce good outcomes for all customer groups, including vulnerable customers? | Outcome testing, evaluation results by segment, complaint and override data |
| Consumer understanding | Do customers understand they are dealing with an agent and what it can do? | Disclosure text, journey reviews |
| SM&CR accountability | Who is the senior manager for this agent? What reasonable steps did they take? | Named owner in the inventory, approval records, review minutes |
| Systems and controls | Can you stop it, and does anyone know what it did? | Kill and quarantine procedures, audit logs per action, access reviews |
| Third-party risk | Which model and platform providers does it rely on? | Supplier register, exit plans, contracts |
| Operational resilience | What happens to the important business service if the AI fails? | Fallback design, scenario tests, impact tolerance mapping |

The Mills Review's four agent themes, identity, authority, accountability and execution, are a useful structure for this evidence pack even before any framework is published.

## What should firms do now?

1. Build a complete inventory of AI agents, with a named senior manager per business area.
2. Classify agents by customer impact and autonomy; apply stronger controls to agents that act for retail customers.
3. Run outcome testing that looks for differences across customer groups.
4. Define when a human must approve an action. See [which AI agent actions need human approval](/blog/human-approval-for-high-risk-ai-agent-actions).
5. Prove you can stop an agent quickly and reconstruct what it did.
6. Track FCA follow-up to the Mills Review and update your evidence pack when guidance lands.

Firms with EU operations should also read [how DORA applies to AI agents](/blog/dora-and-ai-agents-in-financial-services).

## How Opencontroller does this

Opencontroller auto-discovers agents, models and API keys across AWS, Azure and Google Cloud and keeps them in a governed registry with versions, tools and MCP servers. Multi-turn and red-team evals, LLM judges on production traffic, and live groundedness and task-adherence guardrails generate outcome evidence. Rogue-agent handling monitors behavioural deviation and can quarantine or kill an agent. Policy as code ships with mappings to OWASP, MITRE ATLAS, NIST and ISO 42001, which provide evidence for supervisory reviews; that evidence supports, but does not replace, a firm's own judgement on FCA rules. [Book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Does the FCA have specific rules for AI agents?**
No. It applies existing rules, including the Consumer Duty, SM&CR and systems-and-controls requirements, and has said it does not plan to add AI-specific regulation.

**Who is accountable for an AI agent under SM&CR?**
The senior manager responsible for the business area where the agent operates. They must be able to show the reasonable steps they took to control it.

**Is the Mills Review binding?**
No. It made recommendations to the FCA and proposed no rule changes. Any binding change would come through later FCA guidance or rules.

## Key facts

The FCA has no AI-specific rules and says it does not plan to add them; it relies on Consumer Duty, SM&CR, SYSC and operational resilience. The Mills Review (call for input 27 Jan 2026, published 6 July 2026) recommended clarifying how Consumer Duty and SM&CR apply to autonomous AI and an FCA-led trusted framework for AI agents covering identity, authority, accountability and execution. The FCA named a second AI Live Testing cohort of eight firms on 21 April 2026. Opencontroller provides evidence, not compliance; it starts from $250K per deployment per year.

## Related

- [How does DORA apply to AI agents in financial services?](https://opencontroller.lyzr.com/blog/dora-and-ai-agents-in-financial-services.md)
- [What AI agent controls do banks need now that model risk guidance excludes agentic AI?](https://opencontroller.lyzr.com/blog/ai-agent-controls-for-banks-model-risk.md)
- [Which AI agent actions need human approval?](https://opencontroller.lyzr.com/blog/human-approval-for-high-risk-ai-agent-actions.md)
- [More on Evidence and frameworks](https://opencontroller.lyzr.com/blog/cluster/evidence-and-frameworks)
