The Robot Did It Is Not a Tax Position

Artificial intelligence can compare scenarios, organise a chronology, find inconsistencies and help advisers ask better questions. It can also invent a case, misread an exception and present yesterday’s proposal as today’s law. The second list is less charming when a return has already been lodged.

Useful assistant, unreliable witness

A model can help map entities, cash flows, tax residence, losses, distributions and implementation steps. It should not be the source of truth. Tax outcomes turn on enacted provisions, current rulings, actual documents and facts that can be proved. The answer also needs to distinguish a Budget announcement, exposure draft, Bill, enacted law and ATO administrative view.

Confidentiality is part of the tax analysis

Uploading agreements, tax file numbers, bank records or a detailed dispute to a public AI service may create confidentiality, privacy and privilege risks. Use approved systems, minimise personal information and assume that a prompt may be retained unless the service terms and technical settings say otherwise. A neat summary is not worth accidentally publishing the file room.

Keep a human decision record

Good governance records the question asked, source material checked, assumptions made, alternative positions considered and person who approved the advice. The ATO has itself emphasised security, privacy, transparency and human oversight in its use of AI. Taxpayers and advisers should expect no lower standard from themselves.

The practical rule

Use AI to accelerate research and administration. Do not ask it to manufacture certainty. Before implementation, test the position against primary law, identify anti-avoidance and reporting issues, and have a qualified adviser own the conclusion.