AI in financial services: good practice and areas for improvement
AI Analysis
The FCA has published illustrative good and poor practice on firms’ use of machine learning, generative AI and agentic AI, based on industry engagement, supervisory work, AI Live Testing and the Supercharged Sandbox. It introduces no new rules or deadlines, but gives a clear supervisory signal that firms remain accountable for AI-enabled outcomes and should evidence Consumer Duty compliance, senior accountability, lifecycle governance, testing, explainability, monitoring, escalation and resilience.
Key dates
- 2026-10-19
- The FCA publication was issued as part of its wider AI supervisory and industry-engagement work.
Suggested considerations
- Firms should consider maintaining an inventory of AI use cases, including internally developed, embedded and outsourced tools, with a named accountable senior manager and documented risk classification for each use case.
- Compliance teams may wish to map each consumer-facing AI use case to Consumer Duty outcomes, the target market, foreseeable harms, vulnerable-customer considerations, fair value where relevant, complaints handling and available redress.
- Firms should consider documenting why AI is appropriate for the identified consumer need, which activities remain staff-led, where the system must abstain or escalate, and how those controls have been tested in practice.
- Model and risk functions may wish to test accuracy, bias, error rates, explainability, data quality, drift and robustness across relevant consumer characteristics, including vulnerability, accessibility, digital capability and financial capability.
- Firms should consider testing consumer understanding of AI disclosures and outputs with representative user groups, rather than relying solely on technical model explanations or generic disclaimers.
- Operations and technology teams may wish to implement proportionate live monitoring, clear escalation thresholds, sample-based human review, incident and near-miss processes, fallback arrangements and channel continuity where an AI system fails or produces harmful outputs.
- Firms should consider retaining auditable records of prompts, inputs, outputs, model and vendor versions, filters, approvals, overrides, escalations, complaints and remediation decisions for a period consistent with the applicable record-keeping requirements.
- Outsourcing and third-party risk assessments may wish to address concentration and dependency risks, access to testing and performance information, change control, data use, subcontracting, service continuity and the firm’s ability to challenge or replace an AI provider.
What changed
The publication consolidates the FCA’s expectations under the existing framework, including the Principles for Businesses, Threshold Conditions, SYSC, operational resilience, outsourcing and critical third-party requirements, Consumer Duty, vulnerable-customer guidance, SM&CR and DISP. It identifies good practice such as defining the consumer need and target market before deployment, testing outcomes across consumer characteristics and vulnerability scenarios, setting human escalation and abstention controls, maintaining auditable records of prompts, outputs, model versions and decisions, and monitoring live outcomes for bias, drift and harm. It also identifies weaknesses including deploying AI without a defined consumer problem, excluding vulnerable consumers without alternative support,
Compliance impact
The immediate legal effect is limited because the publication does not create new requirements, but its examples are likely to inform FCA supervision and assessments of compliance with existing obligations. Failure to evidence appropriate consumer outcomes, governance, monitoring, escalation, operational resilience or accountability could expose firms to supervisory intervention, remediation and r
Who is affected
Related regulations
References
AI-generated analysis. May contain errors or omissions — verify with the original FCA source before acting. Full disclaimer.
What the FCA said
To support the safe and responsible adoption of artificial intelligence (AI) in financial services, we share examples from industry engagement, supervisory work and our AI Lab initiatives, including AI Live Testing and the Supercharged Sandbox.
Published by FCA . Read the full notice at the source for the authoritative text.