How can insurers prove staff are trained to oversee AI agents?

Insurers already have the right structure: authority limits. Proof comes from attaching a certificate to each. A claims handler or underwriter may approve an agent's settlement or terms only up to their authority, and only while their certificate is current and checkable at source. The filed record then shows, for any decision, who approved it and that they were qualified.

Why insurers have a head start

Insurers grant authority by person, line and amount: a claims handler may settle motor claims up to a set sum, an underwriter may accept risks up to a set limit. That discipline maps well onto agents. An agent can be given depth and limits on the same lines, and the people above it can approve held actions within their own authority.

What insurers often lack is the link between authority and competence on agent work. A handler with ten years' experience and a high settlement limit may never have been trained on how the claims agent reaches its proposals, where its facts come from, or what it refuses. Their authority shows the insurer trusts their judgement on claims. It does not show they can judge an agent's draft.

What evidence do supervisors and auditors expect?

• Authority tables with certificate requirements. For each line and limit, the certificate an approver needs to approve agent-prepared decisions.
• A certificate register. Number, issuer, expiry and status for every approver, checked at source.
• Training mapped to lines. Life and health underwriting is listed as high-risk under the EU AI Act's Annex III; a motor claims handler and a life underwriter need different content.
• Supervised practice records. Decisions made under review before solo authority on agent work.
• The filed record of decisions. Each agent proposal, the approver, their edits and the outcome.
• Complaint and audit sampling results, showing the insurer checks that approvals were sound.

The AI agent training plan template for insurers sets out how to align training with authority tables.

What should claims and underwriting staff be trained on?

1. Provenance in claims files. A policy wording from the insurer's own library is sealed. A repairer's estimate is signed. A claimant's account is declared. A photo from social media is external. An agent may cite all of them in a proposal.
2. Fraud signals the agent surfaces but cannot decide on. Approvers should treat flags as questions to answer, not noise.
3. Fair outcomes. A settlement can be policy-correct and still leave a vulnerable customer worse off than they should be.
4. Held, refused and withheld actions. A declined claim drafted by the agent is held for a person; a payment to a party not on the claim is refused. Staff must not complete a refused action by hand.

An example from a claims audit

A home insurer's claims agent drafts settlement proposals at depth 2 and settles small, clear claims at depth 3 within a limit of 1,000. Internal audit samples 30 settlements between 1,000 and 5,000, each approved by a handler.

Each record shows the handler's name, their authority limit and their certificate status on the day. Twenty-nine are clean. In one, the handler approved a proposal that relied on a contractor's quote attached by the claimant, which the agent had flagged as unverified. The handler was certified, so the issue is judgement, not competence on paper. The audit finding leads to a refresh briefing on external documents for the whole team, logged against each handler.

Which controls make the evidence reliable?

• Authority and certificate checked together before an approval is accepted.
• Automatic suspension of agent-approval rights on lapse, without touching the person's other authority.
• Line-specific refreshers when an agent's sources or limits change.
• An append-only record of every decision.

Market expectations shape the detail; see BaFin expectations in Germany and MAS FEAT principles. For a comparison, see how banks prove staff are trained. The guide to AI literacy and certification explains the overall model.

Where the Academy fits

Pryme Intelligence Academy offers insurers a free readiness assessment for teams of up to 25 people: the team takes the Foundations test set in its industry, and the report shows scores by concept, the gaps and the first agent to start with. That is a sensible first step for a claims team before certifying approvers.