The Senior Managers and Certification Regime applies in full when AI agents do regulated work. An agent cannot hold a senior management function, be certified or breach a conduct rule, so every outcome it produces lands on people: the senior manager whose area it works in, the certified staff who approve its output, and those who set its bounds.
What does the SM&CR require, in plain words?
The regime exists so that a named individual answers for each part of a regulated firm. It has three working parts, described here in substance.
Senior managers hold specified roles, approved by the regulator, and each has a written statement of what they are responsible for. The firm keeps a map showing how responsibilities are split across them, with no gaps and no overlaps. If something goes wrong in a senior manager's area, the regulator asks what that person did to prevent it.
Certification covers staff below senior management whose roles could cause significant harm to the firm or its customers. The firm itself assesses them as fit and proper, and repeats the assessment. Conduct rules then apply to almost everyone: act with integrity, with due skill, care and diligence, and be open with regulators.
The FCA runs the regime for conduct and the PRA for prudential matters in the firms it supervises. Neither has written a separate AI version, consistent with the UK's principles-based, regulator-led approach to AI.
Who does it apply to?
The regime applies to UK banks, building societies, insurers and most firms authorised by the FCA, with the heaviest duties falling on the largest firms. It applies to people, not systems.
That is the point most AI programmes miss. The regime has no category for software, so an agent sits outside it entirely. What sits inside it are the people who decide that an agent will do a piece of regulated work, the people who check what it does, and the senior manager who answers for the business area where the work happens.
What changes when the work is done by an agent that acts?
Three things change, and each creates a gap if the firm does nothing.
First, the volume of decisions rises while the number of people touching them falls. A senior manager who could once sample a team's work now oversees an agent that may draft thousands of letters a week. Accountability does not dilute with volume; the controls must scale instead.
Second, certified functions can quietly move. Suppose a certified adviser used to write suitability letters and an agent now drafts them. If a certified person still reviews and sends each one, the certified function stays with a person. If the agent sends directly, the firm must ask who is now performing a certified function, and the honest answer may be nobody. Keeping the agent at depth 2, where it drafts and a person sends, preserves the line.
Third, conduct rule breaches become harder to see. A person who widens an agent's permissions without approval, or approves drafts without reading them, may be the one in breach. The firm needs a record that shows who did what, as our piece on why senior managers remain accountable for what agents do argues.
How do you meet the regime when agents do regulated work?
Start from your responsibilities map, not from the agent inventory. Every agent should appear under the senior manager whose area it serves, and the deciding question is covered in detail in which senior manager owns an AI agent.
| Requirement | Control | Evidence it leaves |
|---|---|---|
| Named senior manager for each area | Each agent assigned to one senior manager by business area | Responsibilities map and agent register that agree line by line |
| Senior manager can show what they did | Agent bounds, certification results and exceptions reported to that person on a set cycle | Reports received, decisions taken, minutes |
| Certified functions stay with certified people | Agents that touch certified work drafted at depth 2; a certified person approves | Approval records naming the certified individual |
| Staff act with due skill, care and diligence | Approvers trained on the agent's limits and failure modes | Training records and review quality checks |
| Firm knows and controls its risks | Every action checked against role scope, certificate, sources, the firm's limits and platform guardrails | Filed verdict for each action, naming the bound that decided |
| Changes are controlled | Wider permissions require sign-off from the owning senior manager | Change log with approver and reason |
The control that matters most is the last row of evidence: a verdict for each action. Without it, a senior manager cannot show what the agent was allowed to do on the day something went wrong.
What will the FCA or PRA ask?
Expect questions aimed at named people, not at the technology.
- Which senior manager is responsible for this agent, and where does that appear in their statement of responsibilities?
- What did that person know about the agent's permissions before the incident, and what did they do about it?
- Which outputs reach customers without a person approving them?
- Who is certified to perform the function the agent supports, and how do they review its work?
- How do you detect an approver who signs off drafts without reading them?
- What evidence shows the agent behaved within its bounds on a given day?
If you cannot answer the first question in one sentence, the rest will not go well. The companion piece on who is the accountable senior manager under SM&CR works through contested cases.
What does the regime not cover?
The SM&CR does not tell you how to validate an agent, how to test it or how much autonomy is safe. Those questions sit with model risk management, where PRA SS1/23 applies to banks, and with your own operational risk framework. US-regulated groups will meet a similar board-level expectation in SR 11-7 governance and senior management oversight.
It does not cover the agent vendor either. A provider's staff are not your senior managers, so outsourcing the agent does not outsource the accountability. And it says nothing about data protection, which sits with UK GDPR and the Data Protection Act 2018. Read it alongside our guide to agent governance as a set of bounds.
This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
Where Pryme fits
Pryme records every agent action with a verdict that names the bound that decided it, so a senior manager can see exactly what the agent was allowed to do and why. Agents start at depth 1 and earn wider authority only through Attest certification on the firm's own cases, which gives the owning senior manager a documented basis for each step up.
Questions people ask
Can an AI agent be a certified person?
No. Certification is an assessment of an individual's fitness and propriety. If an agent supports a certified function, a certified person must remain responsible for the output, usually by approving it before it reaches the customer.
Does buying an agent from a vendor move accountability to the vendor?
No. The firm remains responsible for the regulated activity, and the senior manager for that area remains accountable. The vendor relationship becomes one more thing that senior manager must oversee.
Do small firms need to map agents to senior managers?
Yes, in proportion. Smaller firms have fewer senior managers, which often makes the mapping simpler: the person responsible for the business line owns the agents working in it.