What the Duty requires
The FCA set out the Consumer Duty in policy statement PS22/9, published on 27 July 2022. The rules came into force for new and existing products or services open to sale or renewal on 31 July 2023, and for closed products or services on 31 July 2024.
The Duty has three layers.
The Consumer Principle. Principle 12: a firm must act to deliver good outcomes for retail customers.
Three cross-cutting rules. A firm must act in good faith towards retail customers; must avoid causing foreseeable harm to retail customers; and must enable and support retail customers to pursue their financial objectives.
Four outcomes. Products and services; price and value; consumer understanding; and consumer support. The FCA describes them as the key elements of the firm-customer relationship: how firms design, sell and service products, and the key contact points along the customer journey.
Underneath sit the obligations that turn the Duty into evidence. PRIN 2A.9 obliges firms to monitor the outcomes retail customers are experiencing from their products. PRIN 2A.8.3R requires a report to the governing body setting out the results of that monitoring and any actions required, and PRIN 2A.8.4R requires the governing body to review and approve that report at least annually, confirm it is satisfied the firm is complying with the Duty, and assess whether the firm’s strategy is consistent with it.
Why agents are inside the Duty
The Duty applies to the firm and to outcomes. It does not distinguish between a person, a script and an agent as the thing that produced the outcome.
Agents are moving into exactly the places the four outcomes describe. A support agent that answers a customer’s question is a consumer-support contact point. An assistant that explains a product’s fees is a consumer-understanding contact point. An agent that carries out a customer’s instruction, moving money, changing a policy, cancelling a product, produces a product-and-service outcome directly.
Two of the cross-cutting rules bite hardest. Avoiding foreseeable harm is a duty to identify and mitigate risks before they materialise, which for an agent acting at machine speed means before the action, not in the complaint that follows. And enabling people to pursue their objectives includes not putting barriers in the way of switching, cancelling or getting help, which an agent can do silently and at scale if nothing constrains it.
What the monitoring obligation asks of an agent estate
PRIN 2A.9 is the requirement that turns agent governance into a Consumer Duty question. A firm must be able to say what outcomes people are getting, and the report to the governing body has to rest on that monitoring.
For a human channel the firm has call recordings, case notes and complaints data. For an agent channel most firms have application logs that show inputs and outputs. The decisions in between, which action, on which system, on whose instruction and under what authority, are usually missing. That is the gap between a firm that can assert its agents deliver good outcomes and one that can show it.
Closing that gap takes a record of every agent action on a customer’s behalf, tied to the customer, the agent, the authority it acted under, the decision and the outcome, kept in a form that has not been altered since it was written. That record is what monitoring under 2A.9 and the report under 2A.8.3R can be built on.
Foreseeable harm and the moment of action
The Duty’s language on foreseeable harm assumes the firm can intervene. For an agent, intervention has to happen on the call.
High-risk actions on a customer’s behalf, money movement above a threshold, closure of a product, a change that removes a protection, should wait for a named person to approve them, inline, with the approval recorded alongside the action. An agent that starts behaving out of pattern should be stoppable immediately, along with anything it set in motion, and the stop should hold without waiting for a credential to expire. And an agent should only be able to reach the systems and take the actions its role requires, so a fault in one agent does not become harm across the estate.
None of these is a Consumer Duty rule. Each is what “avoid causing foreseeable harm” has to mean when the actor is software.
What this page does not claim
The FCA has not published rules specific to AI agents under the Duty, and this page does not say it has. It maps the existing Principle, rules and outcomes to agent controls where the mapping is plain. Whether a particular agent’s activity falls within the Duty, and what good outcomes look like for the people you serve, are questions for your compliance function. This page is not legal advice.
The FCA has said it does not plan to introduce extra regulations for AI and relies on existing frameworks, the Consumer Duty among them. Its AI Update of 2024 sets out how existing rules apply. In January 2026 it launched the Mills Review, a longer look at how AI, including agentic AI, will change retail financial services.
Where KPATH sits
KPATH is the inline enforcement layer for enterprise AI agents. Every governed call passes through its Policy Enforcement Point before the payload reaches the target. When an agent acts for a customer, the call carries the agent’s identity and the customer’s, and policy can insist the person is verified. High-risk actions wait for a named approver, inline. Any agent or a whole chain can be stopped and stays stopped. Every call, decision and outcome is written to a signed, tamper-evident record streamed to your SIEM and verifiable by an auditor without trusting KPATH.
That record is the material for monitoring under PRIN 2A.9 and for the governing body report under PRIN 2A.8. KPATH governs the request envelope, identity, target, action, size, delegation chain, and never reads the conversation, so what the person said and what the agent replied are not held in KPATH’s record.
Start in monitor mode: observe only, enforce no policy, rewrite no agents. See which agents touch customer journeys and what they reach, then set policy and flip to enforce by repointing egress.