Can India's Tax Department See Your ChatGPT Chats?

Guides · By Arthi Rajendran, The Arthi AI Collective · Published · Updated

In the UK this week, tax advisers started asking a question that Indian professionals will recognise: if you used ChatGPT, Claude or Copilot to work out a tax figure, could that chat later be pulled into an investigation?

The short, careful answer is this. In Britain, a proposed change to HMRC's information powers could bring those chats into scope if they were used to produce a tax calculation. It is not law yet, and it is not a licence to read everyone's private history. In India, the government has already said that the digital-search rules in the Income-tax Act, 2025, apply during authorised searches, not as routine AI scrutiny of your chats.

The practical lesson is the same in both places. Your AI chat is a working paper, not a diary. Keep a clean trail. Never paste in anything you would not show an auditor.

This article is not legal advice or tax advice. It explains what official documents and news reports say. For your own situation, please check with a chartered accountant.

What the UK is actually proposing

The UK government has published plans to reform HMRC's civil information and inspection powers under Schedule 36 of the Finance Act 2008, and to update section 114 of the same Act, which deals with computer records. The changes would take effect on or after Royal Assent to Finance Bill 2026-27. They are not in force today.

The official policy paper, updated on 3 November 2026, does not name ChatGPT. What it does say is that section 114 would be updated so that, where HMRC already has the legal right to request information, it can obtain and understand relevant information held in digital systems. That includes "information about how relevant tax figures have been produced including where they are generated using software or automated processes."

The paper is also clear about what this is not. The changes "do not create a new standalone information power." Section 114 would remain a supporting rule, used only where HMRC already has a lawful basis under Schedule 36. Existing safeguards would continue, including the condition that information or documents are reasonably required to check a tax position.

Third parties and intermediaries, including software providers, could also be asked for relevant information they hold or manage in digital form, including what is needed to understand how a tax figure was produced.

That official wording is enough to explain why tax advisers have started talking about chatbot history. Nimesh Shah, chief executive of Blick Rothenberg, told reporters the significant issue was HMRC's plan to obtain information about how tax figures had been produced using software or automated processes. As Eastern Eye reported, he said this raised questions about whether prompt histories, AI-generated working papers, chatbot conversations used to prepare tax calculations, and audit trails within AI agents could become disclosable.

Seb Maley, chief executive of tax insurance provider Qdos, put the same idea more bluntly, as quoted by Eastern Eye: "As more taxpayers and businesses put their trust in AI to help manage their tax affairs, HMRC is making sure it isn't left behind. It wants to know not just what tax figure a system produces, but how it arrived at that number."

An HMRC spokesperson said, as reported by iNews and Eastern Eye: "This is about updating our existing information powers to reflect how tax information is now stored and processed digitally. It is not about going through everyone's use of large language models. We can only request information where we already have a lawful basis to do so."

Privacy campaigners are not relaxed. Anne Strickland, a researcher at the TaxPayers' Alliance, called forcing taxpayers to hand over private AI conversations "a deeply intrusive overreach by the taxman," as quoted by iNews.

So both things are on the table: a narrower official claim (update the rules for digital records, with existing safeguards) and a wider fear (chat histories used for tax work could be pulled in). You do not need to settle that argument to use AI more carefully.

What this means if you are in India

This UK proposal does not change Indian law. If you have a UK tax position as well as an Indian one, it may matter to you directly, and that is a question for your adviser. For most people reading this in India, the useful comparison is our own rule.

The Income-tax Act, 2025, has been in force from 1 April 2026, according to the government's rollout. Section 247 is the search and seizure provision. In February 2026, after questions in the Lok Sabha, the Finance Ministry clarified that Section 247 makes no reference to AI and does not grant fresh or unfettered powers. The ministry said the provision is restricted to duly authorised search and survey proceedings, and cannot be used for routine scrutiny, general data mining or indiscriminate information gathering.

The Press Information Bureau, as reported by Business Today, said the provision is meant for serious offences such as tax evasion and undisclosed income, and that comparable search and seizure powers already existed under the Income-tax Act, 1961. The 2025 law, in that account, largely continues the earlier framework, with updated language for the digital environment.

That is not the same as "nobody can ever see a digital file." Rahul Charkha, a partner at Economic Laws Practice, told Business Today that the Act recognises "virtual digital space," which he said is defined broadly to include email and social media accounts, cloud and remote servers, online banking and trading accounts, digital wallets, crypto-asset holdings and similar platforms. During a duly authorised search or survey, he said, officers may examine electronic records, data or communications in those systems. He added that this could encompass WhatsApp chats, Gmail inboxes, cloud backups and even data stored on overseas servers.

In plain language: the government has said this is not routine monitoring of your chats. It has not said that electronic records are off limits in an authorised search. If a chat you used to work out tax sits on your laptop, in your email or in a cloud account, and you are ever in that narrower search situation, treat it as a record that could be looked at. I am not saying that will happen to you. I am saying it is unwise to write in a chatbot as if the words disappear.

Treat the chat as a working paper, not a diary

David Stirling, a director and financial adviser at Mint Wealth, told iNews something I would give as a habit, not as a legal rule: "If your chatbot helped calculate your tax bill, assume the workings could one day be scrutinised, so don't type anything you wouldn't fancy explaining to the taxman."

That is the standard. A working paper shows how you got to a number. A diary is where people ramble, vent, or ask unsafe questions. If you used ChatGPT to classify an expense, estimate a capital gain, or draft a note for your CA, that chat is closer to a working paper. Write it as if someone might later ask you to explain it.

You remain responsible for the return you file. Chatbots get tax rules wrong. Zoe Dagless, a financial adviser quoted by iNews, made the same point in the UK context: answers are not always accurate and should be checked against official guidance or with a qualified accountant. The same is true here. Check the Income Tax Department's own materials, or ask a CA, before you rely on a generated answer.

A practical checklist

  1. Use AI to understand, not to decide. Ask it to explain a term, outline a process, or help you prepare questions for your CA. Then verify before you file.
  1. Keep your own trail. The record that matters is your spreadsheet, invoices, bank statements and the workings you would happily hand a CA. Do not let a chat history be the only place a calculation lives.
  1. Never paste what you would not show an auditor. That includes passwords, OTPs, UPI PINs, login details, full statements packed with personal identifiers, client information, and any prompt that sounds like a plan to hide income. If you would not put it in an email to your accountant, do not put it in the chat.
  1. Separate the tax work from everything else. Do not drop a tax calculation into the same thread as family messages, medical questions or private venting. A clean thread is easier to review, and easier to stand behind.
  1. Remember that "paid" is not the same as "private," and private is not the same as "beyond a lawful request." I have written about how to keep ChatGPT, Claude and Gemini chats more private, why a paid Gemini plan is not automatically private, and what "private" does and does not mean. Turning off model training is still a good idea. It does not decide what a tax authority can request where the law allows it.
  1. Keep a copy of work you may need later, and do not treat the chatbot as a filing cabinet. If useful tax workings live only inside ChatGPT, export and back up that history. Then store the important figures in your own records.
  1. Do not hand an AI agent your tax-site, email or bank login so it can "just file it." I wrote a separate checklist on what you should never hand an AI agent. Research is fine. The password and the submit button stay with you.
  1. If you are unsure, ask a CA, not only a chatbot. This is especially true for freelancers, consultants, landlords, people with foreign income, and anyone with a UK as well as an Indian tax position.

Keep a trail you can stand behind

I am not asking you to be afraid of ChatGPT, and I am not asking you to stop using it. I am asking you to write in it the way you would write a working paper: clearly, honestly, and without anything you would be embarrassed to explain later.

The UK can argue about how far a tax authority should go into a chatbot history. India has already tried to draw a narrower line around authorised searches. Your habit can be simpler than either debate. Keep the real workings in your own records. Check the answer before you file. When the question is about your own tax, ask a chartered accountant.

At The Arthi AI Collective, I help students and working professionals use AI as a thinking partner, while keeping the judgement, and the sensitive details, their own. If you want to learn to use these tools usefully and carefully, explore the programs.

This article is not legal advice or tax advice.

Sources

Arthi Rajendran, The Arthi AI Collective

Frequently asked questions

Can HMRC read my ChatGPT history?

Not automatically, and not yet as a new standalone power. The UK Finance Bill 2026-27 would update existing information powers so HMRC can, where it already has a lawful basis, obtain relevant information held in digital systems, including how tax figures were produced using software or automated processes. Advisers say chatbot histories used to prepare a tax calculation could fall in scope. An HMRC spokesperson said it is not about going through everyone's use of large language models. The change takes effect only after Royal Assent. This is not legal advice.

Can the Indian income tax department see my ChatGPT chats?

The government has said Section 247 of the Income-tax Act, 2025, does not refer to AI, does not create blanket powers, and cannot be used for routine assessments or broad data collection. It is confined to duly authorised search and survey operations. During such an authorised search, electronic records and communications in a person's digital systems could be examined. This is not legal advice. Check your own situation with a chartered accountant.

What is Section 247 of the Income-tax Act, 2025?

It is the search and seizure provision in India's new income-tax law, in force from 1 April 2026 according to the government's rollout. In February 2026 the Finance Ministry said it modernises the language for digital records, stays in line with the earlier search framework, and is limited to authorised search and survey. It is not a power to browse everyone's email or AI chats as a matter of routine.

Should I stop using ChatGPT for tax questions?

You do not have to stop. Use it to understand a rule in plain language, then check the official source or ask a chartered accountant before you file. Do not let the chatbot be your only adviser, and do not paste anything you would not be willing to show an auditor.

What should I never paste into an AI tax chat?

Passwords, OTPs, UPI PINs and login details. Full bank statements, complete tax workings with personal identifiers, client data, and any question that sounds like a plan to hide income. If you would not put it in a working paper for your CA, do not put it in the chat.

Work with Arthi

Explore 1:1 AI mentorship

Related posts