Regulatory overview

What this section is about

Lending Agent is a working demonstration of an agentic credit broking journey. The conversation a customer has with the agent is, in regulatory terms, the same regulated activity that a human credit broker performs: arranging a regulated credit agreement under article 36A of the Regulated Activities Order. The mediation layer is new. The activity is not.

This section maps the rules that bear on that activity, with citations to the primary sources, and shows where the demo’s behaviours line up with each rule. It is written for four overlapping audiences: FCA-authorised credit brokers, retailers acting as introducers, lenders deciding whether to participate in waterfalls, and reviewers from the FCA’s innovation services. Each page in this section is a self-contained reference; readers can jump in at the rule that concerns them.

Lending Agent is a research demonstrator, not a regulated production service. Nothing on this site constitutes regulated advice or a recommendation. References to “compliance” describe how the demo’s design surfaces evidence; live deployments require firm-level authorisation, governance, and assurance.

The regulatory perimeter

Consumer credit broking in the UK sits inside several overlapping regimes. The relevant ones for an AI-mediated journey are:

Financial Services and Markets Act 2000 (FSMA): which establishes the FCA’s authorisation and supervisory powers. Credit broking is a regulated activity; firms carrying it on without permission breach the general prohibition in section 19 and commit a criminal offence under section 23.
Consumer Credit Act 1974 (CCA) and its secondary instruments, particularly the Consumer Credit (Disclosure of Information) Regulations 2010 (SI 2010/1013), which prescribe the form of pre-contract credit information (the SECCI) and the timing of its provision. Sections of the CCA that touch the agentic journey directly include section 55 (pre-contract information for regulated agreements), section 56 (antecedent negotiations, which makes the lender liable for representations made by a credit broker), and sections 67-68 (cancellation rights and the cooling-off period). Section 75 (joint and several liability of creditor and supplier for misrepresentation or breach of contract by the supplier) applies to debtor-creditor-supplier agreements with a cash price of more than £100 and not more than £30,000; brokered point-of-sale finance for retail goods will often fall in scope. Section 75A provides a comparable but narrower right for linked credit agreements above the section 75 cash-price threshold.
FCA Handbook: PRIN (Principles for Businesses, including the Consumer Duty in PRIN 2A), CONC (the Consumer Credit sourcebook), SYSC (Senior Management Arrangements, Systems and Controls), and DISP (Dispute Resolution: Complaints).
Consumer Duty under PS22/9 and the non-Handbook guidance in FG22/5, which since 31 July 2023 has imposed a higher and more exacting standard than the previous “treating customers fairly” baseline for retail products in scope.
Vulnerability guidance under FG21/1.
Money Laundering, Terrorist Financing and Transfer of Funds (Information on the Payer) Regulations 2017 (“MLR 2017”), SI 2017/692. Authorised credit brokers are within scope and must apply customer due diligence (CDD), retain records, and operate a Money Laundering Reporting Officer (MLRO) regime. CDD obligations sit upstream of the agentic journey but determine what identity attestations the broker needs before the application can be progressed.
Operational resilience: PS21/3, in force from 31 March 2022, with the deadline for firms to operate within their impact tolerances having passed on 31 March 2025. AI-mediated journeys typically support an “important business service” and inherit the firm’s mapping, scenario testing, and impact-tolerance obligations.
Equality Act 2010: section 19 (indirect discrimination) and section 29 (provision of services) bear directly on AI-mediated decisioning where outputs may correlate with protected characteristics. The ICO’s guidance on fairness, bias and discrimination in AI is the working reference for how data-protection fairness intersects with the Equality Act.
Financial Ombudsman Service (FOS): jurisdiction under DISP 2 over complaints from eligible complainants (private individuals, micro-enterprises, small businesses with turnover below £6.5m, certain charities and trusts). The journey, the audit log, and the replay engine all need to be capable of producing evidence to FOS standards on request. See HM Treasury’s March 2026 Review of the Financial Ombudsman Service for the live reform agenda.
UK GDPR and the Data Protection Act 2018, as amended by the Data (Use and Access) Act 2025 (c. 18). The DUAA repealed Article 22 of the UK GDPR and replaced it with Articles 22A to 22D (a more permissive but safeguard-led regime for solely automated decision-making with legal or similarly significant effects); the data-protection reforms commenced on 5 February 2026. The ICO guidance on AI and data protection and the ICO’s draft updated guidance on automated decision-making remain the canonical references.
FCA AI policy posture: set out in the AI Update of April 2024 and the AI Lab initiative including AI Live Testing under FS25/5 (9 September 2025); the second cohort application window opened on 19 January 2026 and closes on 24 March 2026.
Cross-regulator coordination: the FCA, ICO, CMA and Ofcom coordinate through the Digital Regulation Cooperation Forum (DRCF), whose AI and Digital Hub ran a one-year pilot offering a cross-regulator route for queries from innovators. That pilot is now closed to new queries, but its published case studies record the informal advice it gave. AI-mediated credit broking sits at the intersection of FCA conduct and ICO data-protection remits, exactly the kind of question the Hub was set up to field.

This stack is the perimeter the agent operates inside. The pages that follow drill into each layer.

The same regulated activity, with new evidence affordances

A common framing error is to treat AI-mediated broking as a new regulated activity that needs its own rulebook. The FCA’s stated position is the opposite: existing rules apply, and they apply unchanged. What an agentic surface does change is the texture of the evidence available to a supervisor.

A traditional broking journey produces a SECCI, a recorded call (sometimes), an application form, and a CRM trail. An agentic journey produces a deterministic sequence of disclosed facts, a verbatim record of every utterance the customer saw, a structured eligibility result, a consent timestamp, and (in this demo) a replayable transcript that can be re-scored against a held-out evaluation set. The rule has not moved. The evidence has densified.

That densification has consequences on both sides of the ledger. It can be used to demonstrate Consumer Duty outcomes more rigorously than a sampled call review allows. It can also surface failure modes (drift, off-script behaviour, vulnerable-customer mishandling) that a thinner evidence base would have hidden. The Replay and evidence page sets out the methodology used in the demo and where its limits sit.

How this section is organised

Consumer Duty: the four outcomes and the cross-cutting rules, mapped to demo behaviours.
CONC: the Consumer Credit sourcebook chapters that touch the agent surface.
FCA AI strategy: the AI Update, AI Lab, AI Live Testing, and the supervisory expectations.
Vulnerable customers: FG21/1, the four drivers, and the demo’s vulnerability indicator card.
Principles and SMCR: PRIN, accountability for the agent’s behaviour, and the broker as model owner.
Replay and evidence: the §7.7 protocol, what the audit log proves, and the open methodological questions.

What this section is not

This is not a compliance opinion. It is not a substitute for legal advice from a regulatory law firm. It is not an FCA approval, endorsement, or authorisation. The demo at lending-agent.vercel.app cannot be used to introduce a real customer to a real lender; the lender adapters return scripted decisions for evaluation purposes.