Do underwriters need a certificate to approve AI agent actions?

Yes, for underwriters who accept an agent's risk summary, proposed terms or pricing as the basis for deciding. An underwriter's signature binds the insurer to a risk, and in life and health insurance the assessment itself is high-risk under the EU AI Act. The certificate should prove the underwriter can test the agent's reading of the evidence, not its arithmetic.

Why underwriting raises the bar

Underwriting is judgement on incomplete evidence. A good underwriter knows which gaps matter, which disclosures are suspicious, and when a standard rating would be unfair to the applicant or the book. Agents are well suited to gathering evidence and applying the rating guide. They are poorly suited to noticing what is missing.

Regulation adds weight. Annex III of the EU AI Act lists risk assessment and pricing for life and health insurance as high-risk, and deployers must assign human oversight to people with the necessary competence, training and authority. Even outside that scope, an insurer's supervisor will expect the person accepting a risk to be able to explain the decision.

Who does what on an application?

• Agent, depth 1: reads the application, medical evidence, claims history and external data the insurer subscribes to.
• Agent, depth 2: drafts a risk summary, proposes terms against the rating guide, lists exclusions and drafts requests for further evidence.
• Agent, depth 3 within bounds: issues standard terms on simple risks that fall entirely inside the insurer's auto-accept rules.
• Underwriter: accepts, loads, excludes, postpones or declines anything outside those rules; approves every non-standard term; decides on further evidence.
• Refused: declining an applicant on grounds the insurer has excluded from use, and any decision based solely on external data the applicant has not had a chance to correct.

The underwriter's attention should go to provenance. A doctor's report is signed. The applicant's answers are declared. A data broker's lifestyle score is external. An agent that weighs them equally will produce terms that look reasoned and are not.

What should an underwriter's certificate prove?

• That they can read the agent's risk summary and trace every material fact to its source.
• That they know which decisions the agent may make alone and why the rest are held.
• That they recognise when the rating guide does not fit the case, and that overriding the agent is their job, not a failure.
• That they have a supervised, scored record of real decisions with no missed red line, such as accepting a risk on undisclosed material evidence.
• That the certificate is current, held on a register and checkable at source.

An insurer can keep this on an AI agent certificate register, next to the agent's own certification. The same question for collections agents and IT administrators shows how red lines differ by role.

An example on a protection policy

An applicant for income protection declares no back problems. The agent's summary finds no relevant medical history and proposes standard terms. It notes, in passing, a physiotherapy claim on a previous policy two years earlier, recorded in the insurer's own claims system.

The certified underwriter sees that the claims record is sealed, the declaration is declared, and the two conflict. They hold the case, request the physiotherapy notes, and later apply a back exclusion. The agent had found the fact. It had not treated the conflict between sources as decisive, because nothing in its brief told it to. The underwriter then asks for the brief to be updated.

Which controls make underwriting approval safe?

• Authority limits by line and sum insured, each tied to a current certificate.
• A supervised period before solo authority. See how to run a two-week practicum for underwriters, logged in the practicum log template for underwriters.
• Peer review sampling of accepted cases, focusing on conflicts between sources.
• Every decision filed, with the agent's proposal and the underwriter's changes, so a complaint or audit can be answered.

The guide to AI literacy and certification covers how this fits with literacy for claims and distribution staff.

Where the Academy fits

Pryme Intelligence Academy trains underwriters who review agent proposals through its Operator track. The written exam is 40 questions in 60 minutes; the practicum is two weeks of real review decisions signed off by a Pryme assessor. Senior underwriters who design rating agents may go on to the Builder track, whose capstone is an agent certified in Attest to depth 2 or higher.