Yes, for every risk they bind. An AI agent can pre-fill the submission, check it against appetite and draft a price, but binding cover commits the insurer's capital, and the underwriter who binds owns that commitment within their authority. Gaps in data the agent never had, or rules it was given, are shared with whoever owned them. The record shows the split.
Why is binding the moment accountability attaches?
Underwriting decisions last. A risk bound today sits on the book for a year or more, and a mispriced portfolio takes time to show itself. An agent speeds up the work: it reads broker submissions, extracts exposures, enriches them with property or company data, checks the risk against appetite, and drafts terms and a price.
Binding is held for an underwriter with the right authority. For some lines, the stakes are formally recognised: the EU AI Act lists risk assessment and pricing for life and health insurance among its high-risk uses. Even outside those lines, an insurer will want to show that a person with authority looked at every bound risk.
What may the agent do, and what stays with the underwriter?
- The agent may: extract exposures from submissions, enrich them with data from approved sources, run the rating model, check appetite and referral triggers, and draft terms and a quote.
- Held for the underwriter: any bind, any risk that hits a referral trigger, and any quote where a key exposure field is empty or declared only.
- Refused: binding outside the underwriter's authority, and quoting a class the insurer has excluded.
- The underwriter keeps: judging the quality of the risk, negotiating terms with the broker, and deciding exceptions.
What does the record show about who decided?
Every exposure carries its provenance. The insurer's own claims history is sealed. Property data from an approved provider is signed. Figures in the broker's submission are declared until confirmed. The record shows which fields drove the price, which were empty, what the underwriter changed and who bound the risk.
If the agent flagged that a property's flood exposure was declared by the broker as "none" while the approved flood data was unavailable, and the underwriter bound without asking, the decision is theirs.
An example: the missing flood check
Take a commercial property insurer. A broker submits a warehouse with a stated value of 2,000,000 and declares no flood history. The approved flood data source returns nothing for the postcode because of an outage. The agent drafts a quote, marks flood exposure as declared only, and holds it for referral.
The underwriter, keen to meet the broker's deadline, binds at the drafted price. The warehouse floods in the first winter. The record shows the outage, the declared field, the referral and the bind. The underwriter owns the bind. The insurer owns a process where a data outage did not block binding outright.
Which controls fit an underwriting team?
| Control | Evidence it leaves |
|---|---|
| Missing key exposure data blocks binding | Hold reason filed per quote |
| Declared-only fields are shown on the bind screen | Screen state captured in the record |
| Authority limits enforced as bounds | Verdict names the authority bound |
| Supervised practice before binding alone | Practicum log per underwriter |
New underwriters can practise through a two-week practicum for underwriters, recorded in a practicum log template for underwriters. Agent certificates lapse after 90 days and rating rules change; see how to prepare underwriters for 90-day recertification.
The same question arises downstream in claims and elsewhere; compare collections agents and AI agent mistakes and IT administrators and AI agent mistakes. The wider topic is in AI literacy and certification for people who work with agents.
Where the Academy fits
The Operator track at Pryme Intelligence Academy prepares underwriters who review held quotes every day, with module tests set in insurance and in their role. The practicum is two weeks of real review decisions signed off by a Pryme assessor, with no missed red line; binding on a declared-only key exposure is one such line.
Questions people ask
Is the underwriter accountable for the rating model's errors?
Not for the model's design. They are accountable for using judgement on its output, especially when the record shows unusual inputs.
Can the agent bind small risks alone?
Only if the insurer has granted that depth on evidence from its own cases and set tight bounds. The people who set those bounds own the outcomes.
What if the broker misrepresented the risk?
If the underwriter relied reasonably on declared information without visible warning signs, they acted properly. The record shows what was declared and by whom.