Professional indemnity and AI: what your insurer expects of you

By Michael Nadalin, Founder, Market Lead · 11 September 2026 · 7 min read

A junior gives you a memo that cites six authorities. You would not file it without first pulling the two it relies on most. Put a generative AI tool in the junior's place and plenty of practitioners skip that step, because the output reads more cleanly than most first-year work. Your professional indemnity insurer cares about exactly that gap.

Australian courts have already received submissions citing authorities that do not exist, and practitioners have been referred to regulators as a result. None of those matters turned on which software was used. They turned on a lawyer putting their name to material they had not checked.

This piece covers what that means for your PI cover: what the insurer is actually insuring, which uses of AI carry the most claim risk, what a defensible verification process looks like, and when a problem needs to be notified.

Your insurer covers your judgment, not the tool

Solicitors' professional indemnity policies in Australia respond to civil liability arising from the legal services you provide. The error might start in an old precedent, a junior's memo or a language model. Either way the liability is yours, and the question for the insurer and any court is whether you took the care of a reasonably competent practitioner. No category of cover makes a machine error somebody else's problem, and AI vendors' terms typically disclaim accuracy and cap their own liability, so do not count on recovering from the vendor.

In practice, your insurer expects what the profession already expects of you. The Australian Solicitors' Conduct Rules require competent and diligent service (rule 4), prohibit deceiving or knowingly or recklessly misleading the court (rule 19), and require proper supervision of legal services provided under your authority (rule 37). Treat an AI draft as work from a fast, confident junior who nobody has supervised and who sometimes makes things up.

Which AI tasks carry the most claim risk

AI use does not all carry the same risk. The risk depends on two things: how far the output travels before a lawyer checks it, and how hard the error is to undo. You can fix a clumsy email. You cannot fix a limitation date once it has passed.

Litigators have a further layer. Supreme Court of New South Wales Practice Note SC Gen 23, which took effect in February 2025, restricts using generative AI to produce the content of affidavits, witness statements and similar evidentiary material. It also requires that citations in written submissions be verified as existing, accurate and relevant. Other courts, including the Supreme Court of Victoria, have published their own guidelines. Check the practice notes of whichever court you are appearing in before AI goes anywhere near the evidence.

What a defensible verification process looks like

If a claim comes in, a claims manager and a defence solicitor will reconstruct what you did. They will not ask whether you used AI. They will ask what you checked and whether you can prove it. A file note made at the time is worth more than any recollection given two years later.

Checking the citation is not the same as checking the proposition. A case that does not exist gets caught quickly by anyone who looks. A real case cited for something it does not decide is harder to spot, and it gets through when the reviewer only confirms that the citation resolves.

Notification, renewal and disclosure

Legal PI is generally written on a claims-made basis, so when you tell the insurer matters as much as the error itself. If you find that advice or a filed document relied on AI output that turns out to be wrong, that may be a circumstance that could give rise to a claim, even though nobody has complained. Section 40(3) of the Insurance Contracts Act 1984 (Cth) lets a claim made after the policy period attach to the policy under which you notified those facts. If you sit on it, you invite a later argument about which policy responds, if any.

Renewal is the other pressure point. Answer any questions about technology use and controls accurately. If the form asks what safeguards you have, a written policy and a real verification habit make a far better answer than a general assurance.

Confidentiality is a separate exposure

Verification deals with accuracy. The second exposure is what you put into the tool in the first place. Rule 9 of the Australian Solicitors' Conduct Rules prohibits disclosing confidential client information outside limited exceptions. Pasting a client's documents into a consumer chatbot whose terms allow training on inputs, or human review of them, can amount to that disclosure, and it raises questions about privilege too.

A data incident may not sit neatly under your PI policy. Some PI wordings exclude or limit cyber events, so find out now whether a leak through an AI tool would fall under PI, a separate cyber policy, or neither. Firms covered by the Privacy Act 1988 (Cth) also need to think about the notifiable data breaches scheme.

The one-page policy your insurer would want to see

Most firms do not need a long AI governance document. They need one page that someone has actually read. It should list the approved tools, the tasks AI may help with, and the tasks it may not, such as generating evidence or supplying authorities that go into a document unchecked. It should set a verification standard and a file note requirement, and explain how supervisors check AI-assisted work by juniors and paralegals. Barristers in sole practice can apply the same headings to their own chambers practice in an afternoon.

Then hold every tool to the same test. An AI research tool built for Australian law should make the verification step faster by showing you the source behind each answer so you can read it yourself. Apply that test to any tool you consider, Legal Brain included. No tool removes the step, because the name on the document is still yours.

Frequently asked questions

Does professional indemnity insurance cover mistakes made using AI?

Generally the policy responds to your liability for the legal services you provided, and using an AI tool does not change the fact that the work is yours. The real question is whether you met the standard of a competent practitioner, and relying on a machine draft without checking it is unlikely to meet that standard. Read your own policy wording for exclusions, particularly around cyber events.

Do I have to tell my PI insurer that I use AI?

You must answer proposal and renewal questions honestly and disclose matters relevant to the insurer's decision to cover you. If the form asks about AI or technology controls, answer accurately. Separately, if you find AI-assisted work was wrong and could lead to a claim, treat it as a potential notifiable circumstance.

Can I use AI to draft an affidavit in NSW?

Supreme Court of NSW Practice Note SC Gen 23 restricts using generative AI to produce the content of affidavits and witness statements, which must reflect the deponent's own knowledge. Read the current practice note, and the practice notes of whichever court you are appearing in, before using AI anywhere near evidence.

What happens if a lawyer cites a fake case generated by AI in Australia?

It engages your duty not to mislead the court, and Australian courts have already referred practitioners to regulators over AI-generated authorities that did not exist. Costs consequences are also possible. Correct the record with the court immediately and speak to your insurer.