HMRC is about to become another everyday place where artificial intelligence works in the background. That does not mean a chatbot will be doing your tax return for you, but it does mean more of the tax office’s checking, matching and fraud-spotting work may be helped by AI systems.
According to the BBC, HM Revenue and Customs has announced a 10-year, £175m deal with the British technology firm Quantexa. The company’s AI-powered software is expected to help HMRC identify fraud, find unintentional mistakes and support customer service staff.
For ordinary UK taxpayers, the useful question is not whether this sounds impressive. It is what changes in practice when a public body uses AI to look across tax records, external data and payment information. The answer is probably less dramatic than the headlines suggest, but still worth understanding.
What HMRC says the system is for
The BBC reports that Quantexa’s technology will combine data held by HMRC with external sources so the tax office can spot possible fraud and fix mistakes more quickly. The work is expected to include identifying hidden networks of people and companies that may be masking fraudulent activity.
That kind of pattern-spotting is one of the areas where AI can be useful. A human caseworker may be looking at one return, one company or one payment at a time. Software can look for connections across much larger sets of information: repeated addresses, linked businesses, unusual payment flows, mismatched references or patterns that suggest something has gone wrong.
The same technology could also help with less dramatic problems. The BBC says Quantexa expects to help HMRC track legitimate payments that have been made under the wrong reference number. Anyone who has dealt with tax admin knows how stressful a missing or misallocated payment can be, even when the money was sent on time.
Why this matters to normal taxpayers
Most people will not see the AI system directly. They may see its effects through faster checks, better matching of payments, more targeted fraud investigations or different questions from HMRC staff. If it works well, it could help reduce some avoidable delays and make it easier to separate genuine mistakes from deliberate abuse.
That matters because HMRC’s service levels have been under pressure. The BBC reports that a Freedom of Information request by the Contentious Tax Group found more than 93,000 complaints about the department in 2024-25, up from just over 70,000 in 2020-21, with poor response times among the main concerns.
AI will not solve all of that. Long waits, confusing letters and hard-to-reach helplines are not fixed by buying software alone. But if better data matching stops some cases from bouncing around the system, it could make a practical difference for people who simply want their account to show the right balance.
Human checks still matter
The most important reassurance in the BBC report is that Quantexa says AI-made automated decisions about taxpayers will still need to be checked by people. Its chief executive, Vishal Marria, said the technology is designed to support human decision-making rather than replace it.
That distinction matters. Tax decisions can affect people directly: bills, refunds, penalties, investigations, payment plans and business cash flow. If an AI system flags something as suspicious, the next step should not be blind trust in the software. It should be a proper human review with a clear trail of why action is being taken.
This is the same lesson that applies to many everyday AI tools. ManyHands recently looked at why AI tools need oversight before being given too much access. In public services, that oversight is even more important because people may have little choice about whether their data is processed.
What to do if you receive a query
If HMRC contacts you about a return, payment or possible mismatch, do not assume the computer must be right. But do not ignore it either. Treat the letter or message as a prompt to gather evidence and check the details calmly.
Start with the basics: the tax year, National Insurance number or Unique Taxpayer Reference, payment reference, dates, amounts and bank records. If you are self-employed or run a small business, keep invoices, receipts, mileage records, payroll records and bank statements organised enough that you can explain a transaction months later.
Be especially careful with payments made close to deadlines, payments made on someone else’s behalf, old reference numbers, company name changes and amended returns. These are exactly the kinds of boring admin details that can create confusion even when nobody has done anything wrong.
If something looks incorrect, use official HMRC contact routes rather than links in unexpected emails or texts. Scammers often copy the language of tax warnings, refunds and penalties. AI may help HMRC find patterns, but it will not stop criminals from pretending to be HMRC.
Keep records before there is a problem
The practical takeaway is simple: make your records easy for a human to understand. AI systems may flag a pattern, but the resolution often depends on documents, dates and context.
For employees, that might mean keeping P60s, P45s, payslips, benefits-in-kind notices and pension contribution records. For self-employed people, it means clear income and expense records, separate business and personal spending where possible, and notes for anything unusual. For families helping elderly relatives, it means knowing where tax letters, pension statements and payment confirmations are stored.
This is not about becoming fearful of AI. It is about accepting that automated checks work best when your own side of the paper trail is tidy. The same applies when using consumer AI for admin: it can help you summarise, draft and organise, but you still need to verify the final facts. ManyHands has also covered why chatbot answers need checking before money changes hands, and the principle is similar here.
The privacy question
Quantexa told the BBC that HMRC data would remain secure and that staff working with the department would be separate from the rest of the business. Marria also said HMRC data would not be taken away from the HMRC environment.
Those details are important, but they do not remove the need for public accountability. When AI is used in government, people should be able to understand the broad purpose of the system, what safeguards exist, how errors can be challenged and who is responsible when something goes wrong.
For taxpayers, the most useful stance is balanced scepticism. AI could help HMRC find fraud, correct honest mistakes and handle records more efficiently. It could also produce false flags, confusing queries or decisions that are hard for ordinary people to challenge if the process is not explained clearly.
What to watch next
The key test will be whether the new system improves real-world service rather than simply adding another layer of automation. Watch for clearer payment matching, fewer repeated requests for information, quicker correction of obvious errors and better explanations when HMRC asks questions.
If those things improve, AI could be quietly useful. If people receive more automated-looking queries without clear reasons or easy routes to correction, confidence will fall quickly.
For now, the sensible response is not panic. Keep your records, check official messages carefully, challenge mistakes through proper channels and remember that an AI flag is not the same thing as proof. In tax, as in the rest of everyday AI, the best systems should help people make better decisions, not make important decisions disappear inside a black box.
