Plenty of Australians asked a chatbot for help with their tax return this year. The Australian Taxation Office now says the answer has no bearing on who carries the risk if the advice turns out to be wrong. A senior ATO lawyer has said that "AI told me" does not count as reasonable care. The ATO is also considering changes to the two rulings that decide when penalties apply, and a draft could appear within weeks.

What the ATO actually said

Rajitha Srikhanta, the ATO's deputy chief tax counsel, set out the position at The Tax Institute's Women in Tax NSW lunch, on a panel about artificial intelligence. Her words, reported on 30 October, were: "it's not reasonable care if you just say AI told me".

That does not mean the ATO wants people to stop using the tools. According to the same report, Srikhanta said the ATO "does not object" to taxpayers using AI. She also called the technology a "great leveller", because it records each step of its reasoning.

What doesn't change is the legal declaration. Every taxpayer signs to say the information in their return is correct. "You sign a declaration to that effect," Srikhanta said.

The rulings that could change

Two long-standing ATO rulings are on the table. MT 2008/1 explains how the ATO applies penalties for false or misleading statements. It covers the three key standards of behaviour: failure to take reasonable care, recklessness and intentional disregard of the law.

MT 2008/2 deals with shortfall penalties where a taxpayer has taken a position that is not reasonably arguable.

The ATO is considering draft updates to both documents to deal with taxpayers' use of AI. A fact sheet on the "safe and responsible use of AI" is also planned.

The timing is still uncertain. Srikhanta said the draft could come out "around early December". That was her estimate, not a formal ATO commitment, and no firm date has been published.

Until a draft appears, nobody outside the ATO knows exactly how it will be worded. What is clear is the direction: using a tool does not shift responsibility for what goes in the return.

How shortfall penalties are calculated

The penalty framework itself is not new. It sits in Schedule 1 of the Taxation Administration Act 1953, and section 284-90 sets the base penalty rates for false or misleading statements that cause a tax shortfall.

  • Failure to take reasonable care: 25 per cent of the shortfall amount.
  • Recklessness: 50 per cent.
  • Intentional disregard of the law: 75 per cent.

These are base rates. The final penalty can be increased or reduced depending on how the taxpayer behaves, including whether they tell the ATO about the error voluntarily.

Hypothetical example: Sam asks a chatbot whether a deduction is allowed, receives a confident yes, and claims it without checking. The ATO later decides the claim was wrong, and the error leaves a $4,000 shortfall in tax.

If the ATO decides Sam failed to take reasonable care, the base penalty is 25 per cent of $4,000, or $1,000. If it decides the behaviour was reckless, the base penalty is 50 per cent, or $2,000. A finding of intentional disregard would mean a 75 per cent base penalty, or $3,000. The $4,000 of tax still has to be paid in each case.

This example is illustrative only. The ATO weighs each person's circumstances, and the outcome in a real case could be different.

Why Chatbots trip people up

Earlier warnings from the ATO show where things can go wrong. On 3 October 2025, a commercial survey of 1,000 Australians found that nearly a third had used AI tools to complete their tax returns. The survey was commercial research, not official data.

In the same report, ATO assistant commissioner Robert Thomson warned: "AI can sound really confident in its answer even when it's wrong."

One example shows how easily outdated or mistimed information slips in. A $1,000 standard deduction for work-related expenses has been legislated, receiving Royal Assent on 26 June 2026, but it starts from 1 July 2026. The ATO's own Tax Time 2026 guidance makes clear it does not apply to 2025–26 returns. A tool that blurs those dates could suggest a claim that isn't yet available.

The ATO's Tax Time 2026 guidance also says myth-busting messages appear in myTax to catch common errors.

In July, the ATO had warned taxpayers against relying on AI and "finfluencers" for tax advice. Assistant commissioner Anita Challen said: "Remember you are making a legal declaration that the information in your tax return is true and correct."

Using AI and the risk of an error

Nothing in the ATO's comments bans these tools. The issue is what happens after the answer comes back. Some general points follow from the ATO's position.

A chatbot's answer is a starting point, not a final ruling, and it can be compared with the ATO's own guidance, particularly on start dates and eligibility.

Records of the evidence behind each claim matter more than the chat history. Receipts and logbooks still do the heavy lifting if a claim is questioned.

If something is complex or high-value, a registered tax agent remains an option. Unlike a chatbot, an agent is regulated.

If you discover an error after lodging, the ATO generally treats a voluntary disclosure more favourably than one it finds itself.

Key dates and what to watch

  • Early December 2026: Srikhanta's estimate for draft updates to MT 2008/1 and MT 2008/2. Treat this as possible, not confirmed.
  • With or after the drafts: a planned ATO fact sheet on the safe and responsible use of AI.
  • Comment period: draft ATO guidance is usually open for public comment before it is finalised. Watch for how long this window runs.
  • 1 July 2026 onwards: the $1,000 standard deduction applies from the 2026–27 income year. That is a date AI tools can easily mix up.

The main point for taxpayers is that the obligation in the declaration stays with the person who signs it. Whatever the ATO's final wording, using AI does not change whose name goes on the return.