Yes, for simple risks inside a written appetite. An AI agent can gather the proposal, run the approved underwriting rules and bind standard personal-lines cover at depth 3. Referrals, declines, non-standard terms and complex commercial or life risks stay with an underwriter, because once cover is bound the insurer is on risk and a loss can arrive the next morning.
What does binding cover commit the insurer to?
Binding is the moment the insurer accepts the risk. From then on, if the insured event happens, the insurer pays. A policy can usually be cancelled later, and many products give the customer a cooling-off period. But cancellation does not reach back past a loss that has already occurred.
That is what makes underwriting different from quoting. A wrong quote costs a refund. A wrong acceptance can cost a full claim on a risk the insurer would never have written. And a wrong decline leaves a customer without cover they needed, perhaps for a home or a business that cannot trade uninsured.
Both errors are hard to reverse, which is the test set out in why autonomy in regulated industries is a question about reversal. The safe zone is where the rules are clear, the risk is standard and the cost of a wrong acceptance is bounded.
Which underwriting steps can the agent take?
| Step | The agent | Stays with an underwriter |
|---|---|---|
| Gathering the proposal | Collects answers, documents and third-party data | Nothing routine |
| Checking disclosures | Compares answers with external data and flags mismatches | Deciding whether a mismatch matters |
| Applying appetite rules | Tests the risk against each written rule | Interpreting rules that do not fit the case |
| Pricing | Runs the approved rating model | Overrides |
| Binding standard risks | Binds at depth 3 when every rule passes | Nothing routine |
| Referrals | Drafts the referral with what triggered it | The decision |
| Declining cover | Never | Every decline |
| Non-standard terms or exclusions | Drafts the proposed wording | Agreeing it |
Pricing is its own question, covered in whether an AI agent can set insurance premiums. Underwriting decides whether to take the risk and on what terms; pricing decides what to charge for it.
The agent can also answer customers' questions about the cover they are buying, using the approved wording, as described in answering insurance policy questions with an AI agent.
A worked example
Take an insurer writing home contents cover for tenants. The appetite rules accept flats and houses in standard construction, contents up to a set sum insured, and no more than one claim in the past five years.
A proposal arrives for a flat, contents of 25,000 and no claims. Every rule passes. The agent binds the cover at depth 3 and issues the documents.
A second proposal declares no claims, but the agent's check of the claims data shared across insurers shows a theft claim two years ago. The rule would still pass with one claim, but the disclosure does not match. The agent does not bind and does not decline. It refers the case to an underwriter with both pieces of evidence. The underwriter speaks to the applicant, who forgot the claim, corrects the proposal, and accepts the risk.
What controls make it safe?
- Appetite as written bounds. Every rule the agent applies is written down and owned by underwriting. If a case does not fit a rule, it is referred, never interpreted.
- Declines are always held. Refusing cover is outside the agent's scope at every depth.
- Disclosures are declared, not verified. What the applicant says is a declared source. The agent checks it against independent data where it can, and treats any gap as a referral.
- A reversal window. Define what happens if an agent-bound policy is later found to be outside appetite, and how long the insurer has to act. The reversal window policy template sets this out.
- A written policy. Record which products, limits and rules the agent may bind under in your AI agent policy, so the scope is approved before it is used.
The EU AI Act lists risk assessment and pricing in life and health insurance among its high-risk uses. Underwriting those products with an agent brings duties on risk management, logging and human oversight, and is better kept with an underwriter.
Where Pryme fits
In Pryme, binding a standard risk and drafting a referral are separate capabilities with separate depths. Attest certifies each one by replaying the insurer's own past proposals and scoring every result against a floor. The binding capability stays at depth 3 only while its certificate holds; certificates lapse after 90 days, and a run that clears no depth drops the agent to depth 1 the same day.
Questions people ask
Can the agent decline a risk that is plainly outside appetite?
It can stop the application and draft the decline with the rule that failed. A person sends it, because applicants sometimes give wrong answers by mistake, and a decline may need to be disclosed to future insurers.
Can the agent underwrite commercial risks?
For small, standard packages with clear rules, the same pattern can apply. For larger or unusual commercial risks, the agent prepares the submission and the underwriter decides.
Who is accountable for an agent-bound policy?
The underwriting manager who owns the capability. The record shows which rules passed, which data was checked and under which bound the policy was written.