AI and your professional indemnity cover

One of the quietest worries about AI is not whether it works. It is whether using it could come back on you: your regulator, your insurer, a claim. Here is what is actually known, in plain terms, and how to keep your AI use defensible. This is general information, not legal or insurance advice.

You stay responsible. The AI does not take the blame.

Every UK regulator that has spoken on this says the same thing: using AI does not move your professional responsibility onto the tool. The Solicitors Regulation Authority puts it plainly, that adopting AI does not diminish a solicitor's obligations, that authorised individuals remain accountable for the work, and that outputs must be reviewed and supervised. The duty to keep client affairs confidential and to hold client information securely continues to apply whatever technology you use.

The same principle runs through the other professions. If you are an accountant, a doctor or in HR, you remain answerable for the advice and the documents that leave your desk, whether a junior, a template or a model produced the first draft. AI changes how the work is done. It does not change who is responsible for it.

Where the liability actually sits

The risk that matters for a claim is rarely "the AI wrote something wrong" on its own, because you are reviewing the output. It is what happens to the confidential material you put in. Paste a client file, a patient letter or an employee grievance into a consumer AI tool and, on most of their consumer terms, that text is stored, may be used for training, and can sit under a legal regime outside the UK. A UK tribunal has held that putting confidential documents into a public AI tool can waive legal professional privilege and trigger a duty to notify the regulator and the ICO. That is a confidentiality event, and confidentiality events are exactly what professional indemnity claims and regulatory findings are made of.

You can see, per tool, what that actually means with our AI training checker and the fuller data teardown, or test jurisdiction with the CLOUD Act checker.

What your insurer expects

A professional indemnity policy is priced on the risk you present. Under the Insurance Act 2015, a commercial policyholder owes a duty of fair presentation of that risk, which is why a material change in how you work can be the kind of thing an insurer expects to hear about. Whether your particular use of AI is material is a question only your insurer or broker can answer, so the safe move is to ask rather than assume. Many brokers now advise disclosing material AI use so that cover is not disputed after a claim.

What insurers and regulators reward is the same short list, and it is worth keeping the evidence for all of it:

• A written AI policy: which tools are allowed, for what, and what may never be pasted in. Our note on what goes in a practice AI policy is a starting point.
• Human review of every output before it reaches a client. Documented human oversight is widely regarded as the strongest defence on a negligence claim.
• An audit trail you can actually produce: what was generated, when, by whom, and that it was checked.
• A defensible tool choice: one that does not expose the confidential material in the first place.

How Hush is built for that record

Hush is designed so the defensible choice is the easy one. It keeps no conversations or documents, so there is no stored copy to leak or be compelled. It never trains on what you type. It runs on hardware we own, with no US parent and no third-party model API in the path, so there is no outside company that could be ordered to hand your data over. And it gives you an audit log you own, the written evidence that the work was generated privately and reviewed by a person. The professional judgment stays yours; the paper trail that protects it is produced for you.