A taxpayer asks a chatbot how to claim a home office deduction, gets a confident answer, files a return based on it, and later discovers the advice was wrong. That scenario, repeated across millions of queries, is exactly what Britain’s tax authority is now trying to get ahead of. According to Bloomberg, HMRC is preparing to tackle rogue advice generated by artificial intelligence, a signal that regulators increasingly see consumer-facing AI as something that needs supervision rather than just enthusiasm.
The timing is not accidental. Over the past two years, millions of people have started treating general-purpose chatbots as a free substitute for accountants, lawyers, and financial advisers. The tools are fluent, fast, and eager to please. The problem is that fluency is not the same as accuracy. Language models generate plausible text based on patterns, and tax law is one of the least forgiving domains for a plausible-sounding error. A miscalculated allowance or a misunderstood deadline can mean penalties, back payments, or lost relief.
HMRC’s interest reflects a broader shift in how governments view AI. For a while, the regulatory conversation focused on grand questions about safety and existential risk. Now the focus is narrowing to practical harms: what happens when automated systems give ordinary people bad advice on matters that carry legal and financial consequences. Tax is near the top of that list because everyone has to interact with the system, and the penalties for getting it wrong land on the individual, not the chatbot.
There is a genuine tension here. On one hand, AI assistants can democratize access to basic information. Someone who cannot afford a consultation with a tax professional might get useful orientation from a well-designed tool. On the other hand, the same tool can produce authoritative-sounding nonsense with no obvious warning signs. The burden currently falls on users to verify everything, and most people are not equipped to do that in a field as dense as tax law.
The enforcement question is delicate. HMRC cannot easily regulate every model that might answer a tax question. It can, however, scrutinize the companies deploying these tools, issue guidance on what counts as responsible use, and pursue cases where AI-generated advice crosses into unregulated professional services. The authority has long policed unqualified tax agents and aggressive avoidance schemes, so extending that remit to automated advice is a natural, if technically awkward, next step.
For the companies building and deploying these systems, the message is that disclaimers alone will not be enough. A line at the bottom of a chat window saying the information is not professional advice offers thin protection when the interface itself invites users to rely on it. Expect more pressure toward guardrails: refusing to answer certain questions, routing users to official sources, or building verification steps into the response. The firms that treat compliance as a product feature rather than an afterthought will likely fare better than those that wait for enforcement.
There is also a lesson for the public. The appeal of asking a chatbot about taxes is obvious, but the technology is not a licensed adviser and does not carry professional liability. Treating it as a starting point for research, rather than a final authority, remains the sensible posture. Cross-checking against official guidance and, where the stakes are high, consulting a qualified professional is still the safest route.
HMRC’s move is unlikely to be the last of its kind. As AI tools become the default first stop for questions about money, health, and law, regulators across sectors will face the same problem: how to protect people from bad automated advice without stifling a genuinely useful technology. The UK taxman’s scrutiny is an early test case, and the rest of the world will be watching how it plays out.