FCA authorised · FRN 724952 0117 325 0027Quote & buy →
Aipedia: Tax advisers
AI chatbots can speed up tax research, but they invent or misdescribe cases, so every authority must be checked against a primary source before use. Tribunals have already seen fake tax cases, and advice built on them is likely to be hard to defend.
Part of: AI in Tax advisers
In short
Generative AI tools produce fluent answers on tax law, complete with case names and section numbers that may not exist. In Harber v HMRC (2023) a taxpayer cited nine First-tier Tribunal decisions that turned out to be AI-generated. In Zzaman v HMRC (2025) AI-prepared citations did not support the points made. The PCRT AI guidance (para 3.4) tells members to treat output like a junior colleague's draft and confirm that case law and legislation exist. Advice resting on wrong law is a classic negligence claim, which PI usually covers subject to the wording.
Advisers use chatbots to summarise a relief, find the section that governs a point, list cases on a technical issue or explain an HMRC manual paragraph. The answer arrives in seconds, in confident prose, often with citations. The difficulty is that a language model predicts plausible text. It can produce a case name, a neutral citation and a quotation that look real and are not, or attach a real case to a proposition it does not support.
In Harber v HMRC [2023] UKFTT 1007 (TC) an unrepresented taxpayer appealing a CGT failure-to-notify penalty relied on nine FTT decisions. None were genuine, and the Tribunal found they had been generated by AI (para 21). It said fake citations "promote cynicism" about judicial precedent and waste time and public money (para 24). The appeal was dismissed.
In Zzaman v HMRC [2025] UKFTT 539 (TC), a High Income Child Benefit Charge appeal, the statement of case had been prepared with AI and the cited authorities did not support the propositions. The Tribunal said this "highlights the dangers of reliance on AI tools without human checks" and that for hallucination "There is no reliable way to stop this" (para 29).
Both taxpayers were unrepresented. A professional who made the same error would be judged by a higher standard.
Illustrative scenario: a client asks whether a payment qualifies for a particular relief. The adviser asks a chatbot, which confirms it does and quotes two tribunal decisions. The adviser includes the conclusion in a short email without opening either decision. One case does not exist and the other decided the opposite. HMRC opens an enquiry, denies the relief and charges a penalty. The client says they acted on the email and claims the tax, penalty and costs from the adviser.
Quieter versions are more common: a rate or threshold that has since changed, a repealed provision quoted as current, or an HMRC manual paragraph paraphrased wrongly.
A claim that advice was wrong in law is the core risk that tax advisers' PI is designed for. The CIOT/ATT PII Regulations require cover for all civil liability from tax services (4.1(b)), so the use of AI would not usually be the issue; the question is whether the adviser fell below the standard of a reasonably competent tax adviser. After Harber and the PCRT guidance, failing to check that an authority exists will be hard to defend. Tax that was always payable is not usually a recoverable loss, but penalties, interest and costs may be, depending on the facts and the wording. See AI hallucination liability and tax advice PI claims.
Related guides: AI-drafted tribunal submissions and AI-drafted advice reports. Back to the AI in tax advice hub.
Apex arranges PI for tax advisers. A broker can check how your policy responds to technical advice errors and whether your research process is fairly described to insurers.
Yes. The PCRT AI guidance accepts AI use but says output should be treated like a junior colleague's draft. Use it to find leads and summarise, then verify every case, section, rate and threshold against an official source before advising.
No. Both taxpayers were unrepresented. The tribunals still criticised the use of unchecked AI citations. A professional adviser making the same mistake would be judged against the standard of a reasonably competent tax adviser, which is higher.
Usually PI responds to negligence claims regardless of the tool involved, subject to the policy terms. The claim may be hard to defend, and losses such as penalties and costs may be recoverable, while tax that was always due usually is not.
Tell us about your firm and the work you do. A broker will review how your current PI treats AI-assisted work, or arrange new cover. No fee for a quote. Or call 0117 325 0027.
Start a PI quote Request a callbackApex Insurance Brokers Limited is authorised and regulated by the Financial Conduct Authority. Registered in England and Wales, company number 07014570. This page is general information about how AI is used in the profession and how insurance may respond. It is not legal, regulatory or insurance advice on your circumstances, and it does not guarantee that cover will be available or on what terms.
