What UK professional indemnity insurers actually ask about AI use

The AI liability question that actually costs UK regulated firms money is not "can AI make a mistake". It is "what happens to our professional indemnity cover when it does, and what does our insurer expect us to have already put in place."

The current market position

As of mid-2026, UK insurance brokers who place professional indemnity (PI) cover describe most insurers as not excluding AI-assisted work as a class. The reasoning is straightforward: the professional using an AI tool carries the same civil liability whether the underlying work was AI-drafted or written by hand. A standard professional negligence claim, where AI-assisted advice turns out to be wrong and a client acted on it, is generally treated the same way any other negligence claim would be.

What is changing is not coverage exclusion so much as disclosure expectation. Some insurers are beginning to ask specific questions about AI tool usage on the renewal proposal form, and some newer policy wordings introduce sub-limits or excess loadings specifically for AI-driven claims. Brokers are explicit that the market has not settled on one approach, which means the answer a firm gives at renewal matters more than it did two years ago, not less.

The four scenarios that actually create exposure

Brokers placing this cover describe a consistent pattern across the claims that concern insurers most:

• Unchecked AI-generated advice: a professional uses AI to draft advice and delivers it without review. If the AI produces a factually wrong statement and a client acts on it, this is treated as a standard professional negligence claim.
• AI as sole decision-maker: a firm uses AI to make client-affecting decisions with no human oversight, automated underwriting or algorithmic advice without a person checking the output. This is where insurer scrutiny sharpens noticeably.
• AI training on client data: if an AI provider trains on client-identifiable data without consent, this sits closer to a confidentiality or data-protection exposure than a classic PI claim, and may fall under a cyber policy rather than PI cover, sometimes both, sometimes neither.
• Misattributed AI-assisted work: a professional signs off AI-generated work as their own without appropriate supervision, and a subsequent complaint challenges whether genuine professional judgement was actually applied.

The insurer cares less about the tool and more about the supervision. A firm that can describe its supervision protocol specifically is answering a different, better-priced question than one that describes its AI use in vague terms.

What insurers actually want to see disclosed

Where a renewal proposal asks about AI use, brokers advise answering with specifics rather than generalities: which tools are in use, for which categories of work, what the supervision protocol actually is, and whether a qualified professional reviews AI-generated output before it reaches a client. An underwriter reading a vague answer prices the unknown risk conservatively. An underwriter reading a specific, documented process has something concrete to assess.

The coverage gaps worth checking before a claim forces the question: whether the policy's definition of "professional service" is drafted narrowly enough that AI-assisted advice might sit outside it; whether the wording treats AI use as an implicit delegation the insurer wants the professional to retain responsibility for; whether newer wordings sub-limit or exclude AI-driven automated decisions specifically; and whether an AI-related confidentiality failure would actually be covered by the PI policy, the cyber policy, both, or neither. These are not hypothetical questions for a renewal conversation. They are the specific points brokers say are worth confirming at inception, not discovering at claim time.

Professional-body guidance still matters

For regulated professions, the relevant professional body's own AI guidance shapes how an insurer, and eventually a regulator, will view a firm's conduct. The Solicitors Regulation Authority has published guidance on AI use covering supervision, competence, confidentiality and transparency to the client. Architects' PII adequacy requirements extend to AI-assisted design work. Accountancy bodies including the ICAEW have issued guidance on AI in advisory and audit contexts. The FCA's Consumer Duty applies fully to AI-assisted retail investment advice: using an AI tool does not change what the four Consumer Duty outcomes require. Surveying's regulator, RICS, has begun addressing AI-assisted valuation specifically. Compliance with the relevant body's guidance supports a firm's position when it comes to fair presentation of risk at PI renewal, not just good practice for its own sake.

What this actually costs a firm to get right

None of the practical steps brokers recommend require a large budget: a documented AI use policy naming which tools are used for which work, human-in-the-loop review maintained for anything client-facing, an inventory of AI tools in use, AI use addressed explicitly within a firm's Consumer Duty framework where applicable, AI use disclosed specifically rather than vaguely at renewal, and AI-related complaints or notifications tracked separately for annual review. The real cost is not the documentation itself. It is the gap between firms that can answer an underwriter's questions specifically, and firms that discover at claim time that their policy never clearly said whether it covered this in the first place.