FCA Consumer Duty and AI: Practitioner Q&A for Regulated Firms
How the FCA applies the Consumer Duty to AI: the four outcomes, vulnerable customers, senior manager accountability and the evidence supervisors look for.
This page is for compliance leads, CROs and senior managers in FCA-regulated firms that use AI with customers or plan to. The FCA has said its existing framework, including the Consumer Duty, applies to AI. That makes the useful question one of evidence: what shows how an AI system behaves with real customers, including vulnerable ones and who is accountable for it? The answers below describe the FCA’s published position and point to its sources. Nothing on this page is regulatory or legal advice.
Does the FCA Consumer Duty apply to AI?
Yes. The FCA applies its existing framework to AI and names the Consumer Duty among the rules that apply.
Its AI Update of 22 April 2024 maps the government’s five AI principles onto existing rules: the Principles, the Consumer Duty, SM&CR, SYSC, operational resilience and outsourcing. Its page AI and the FCA: our approach calls that approach principles-based and outcomes-focused. On 19 October 2026 it published good practice and areas for improvement on AI in financial services, tying AI to the Duty’s outcomes, to consumers in vulnerable circumstances and to senior manager accountability.
In the 2026 Bank of England and FCA survey, respondents cited the AI Update (65%) and the Consumer Duty (63%) most often as relevant FCA guidance. See also our Q&A on UK AI regulation by sector.
Does the FCA have specific rules on AI?
No. The FCA has no standalone AI regime and continues to apply existing requirements to AI.
The October 2026 good and poor practice publication creates no new requirements, and its examples are illustrative and not exhaustive. A compliance lead could treat it as optional. We read it as the clearest published view of what supervisors look for when a firm uses AI with customers.
Two developments may add detail. In January 2026 the Treasury Committee recommended that the FCA publish practical guidance by the end of 2026 on how existing consumer protection rules apply to AI; that is a parliamentary recommendation and binds no firm. The Mills Review, published on 6 July 2026, made seven priority recommendations to the FCA Board, which decides which to take forward. Neither pauses the framework the FCA already applies.
How does AI affect the four Consumer Duty outcomes?
AI can bear on all four: products and services, price and value, consumer understanding and consumer support. The FCA’s October 2026 publication groups its Duty examples under products and services, under consumer understanding and under consumer support and vulnerable consumers.
Its good practice examples include defining the consumer need and the market a product is for before building AI. They include deciding which tasks AI should support, automate or abstain from and where it should escalate to a person, which is human oversight set by design. On understanding, they include telling consumers how AI is used and what its limits are, a task that leans on AI explainability.
On price and value, AI fairness becomes the question where automated decision-making sets prices or eligibility. The Duty’s outcome monitoring provisions in PRIN 2A.9 cover whether different groups of customers receive different outcomes.
How should FCA-regulated firms using AI treat vulnerable customers?
The FCA’s position is that its vulnerability guidance, FG21/1, still applies alongside the Duty, and its AI examples caution against two shortcuts.
Its March 2025 multi-firm review concluded that FG21/1 remains appropriate and that it would not revise the guidance or add requirements. The FG21/1 page now directs firms to the Consumer Duty pages for current expectations on management information and good outcomes.
The October 2026 publication lists testing outcomes across consumer characteristics, including vulnerability and digital or financial capability, as good practice. It flags excluding consumers with vulnerability indicators from AI-supported services by default as an area for improvement, so the simplest control is one the FCA questions. It also cautions against over-relying on language models to detect vulnerability. What remains is showing how the system treats vulnerable customers in practice.
Who is accountable for AI under the SM&CR?
Accountability sits with the senior manager whose responsibilities cover the activity, business area or management function where the AI is used. No AI-specific senior management function or Prescribed Responsibility exists.
The AI Update states that any use of AI falls within the scope of an SMF manager’s responsibilities, and the reasonable steps standard applies. In dual-regulated and Enhanced firms, technology systems normally sit with SMF24 and overall risk management with SMF4. In FS2/23, most respondents did not think a new Prescribed Responsibility for AI would help.
Accountability spread across existing SMFs can blur. The FCA’s October 2026 good practice examples include mapping AI responsibilities across SMFs, and we see that map as algorithmic accountability in working form.
The SM&CR Phase 1 reforms in PS26/6 left the allocation of AI accountability unchanged. The Phase 2 consultation papers of 22 October 2026 are proposals only and depend partly on legislation.
How do you evidence good customer outcomes from an AI system?
Firms evidence them by recording how the system behaves with real customers and comparing that record with the outcomes they intended.
The FCA’s good practice examples point the same way: testing outcomes across consumer characteristics and using proportionate bias and accuracy controls. The publication accepts sample testing for smaller firms where large-scale testing is not feasible.
Useful evidence includes test results by customer group, records of escalations to a person and samples of real interactions reviewed against the intended outcome. Firms with model risk management, second-line compliance and internal audit may already hold much of this, and the FCA says firms can build on existing frameworks.
A fair test is whether someone who did not build the system could follow the evidence from customer interaction to outcome. That independence is the core of AI assurance.
How should we monitor AI systems after launch under the Consumer Duty?
The reference points are the Duty’s outcome monitoring provisions in PRIN 2A.9 and its governance provisions in PRIN 2A.8; the FCA has set no separate monitoring regime for AI.
What AI adds is change after launch. Behaviour that passed testing can shift when a model, prompt or data source changes, so we would repeat outcome tests on a schedule and after each change. Where staff review AI outputs, we would also check whether they challenge the system or defer to it, the pattern known as automation bias.
One proposal to watch is CP26/23, published on 29 June 2026. It proposes reporting proportionate to a firm’s role and no standalone Consumer Duty board report where the information is covered elsewhere. These are proposals only: the consultation closed on 18 September 2026, final rules are expected in early 2027 and PRIN 2A as it stands remains the operative text.
What should a regulated firm check before launching a customer-facing AI system?
We suggest four questions, each answered with evidence: who the system is for, what it should and should not do, how it behaves across customer groups and which senior manager owns it.
Each maps to the FCA’s October 2026 good practice examples: defining the consumer need before building; deciding which tasks AI supports, automates or abstains from and where it escalates to a person; testing outcomes across consumer characteristics, including vulnerability; and mapping responsibilities across SMFs.
Check third-party systems with the same care. They made up 39% of reported AI systems in the 2026 Bank of England and FCA survey, up from 33% in 2024, although the Bank cautions that samples differ between editions. The firm remains responsible for outcomes from those tools, so confirm that the contract allows the testing you need.
These checks turn responsible AI principles into records a supervisor can read.
AI Behaviour Verification
AI Behaviour Verification is our ongoing independent verification that AI systems behave as intended, with regular reports that serve as evidence for auditors.