UK War Against Freedom Continues

The Telegraph piece reports that HMRC's information powers are set to expand under the Budget Bill (expected to become law by the end of May). Currently HMRC can demand computer "records" via Schedule 36 notices when checking a taxpayer's position. The new wording broadens this to any "relevant thing" in digital systems and extends the reach to third parties (including AI platforms and software providers).

Tax professionals quoted in the article note that this could bring within scope prompt histories, chatbot conversations, AI-generated working papers and audit trails from tools such as ChatGPT, Copilot or Claude, if those materials formed part of how a tax figure was produced. Failure to comply with a valid notice carries an initial £300 fine (higher for unreported inaccuracies). The change sits alongside other tax-gap measures, including the planned ability (from July 2029) to recover low-value debts directly from bank accounts where taxpayers have repeatedly failed to engage.

HMRC's stated position is that this updates existing powers for the way tax information is now stored and processed digitally; it is not a general trawl of everyone's large-language-model use, and requests still require a lawful basis. Critics (including the Taxpayers' Alliance) call the prospect of demanding private AI conversations a further intrusive expansion of already broad powers. Surveys cited in the coverage show substantial real-world use of generative tools for tax and financial questions by both individuals and businesses, which is the practical backdrop for the change.

In short, the article describes a targeted widening of information-gathering powers in tax inquiries rather than routine mass surveillance of all AI chats. Whether the expanded definition is applied narrowly or more aggressively will depend on how HMRC and the tribunals interpret "relevant" in practice once the Bill is law.

https://www.telegraph.co.uk/money/tax/news/hmrc-to-probe-taxpayers-ai-chat-history-in-latest-crackdown/