Artificial intelligence (AI) is becoming part of everyday tax practice. A practitioner asks an AI tool to draft a tax calculation and prepare the first version of advice for a client. Within seconds, the work is done. 

The draft is clear, well written and persuasive. It is also wrong.

If the tax position later proves to be incorrect, has the taxpayer taken reasonable care? And does the answer change simply because AI was involved?

The short answer is no. AI changes how tax work is performed, but it does not change the legal standards that apply. 

AI is changing the way practitioners work

For many practitioners, AI has already become another research and drafting tool. It can quickly summarise Inland Revenue guidance, produce first drafts of calculations, explain unfamiliar concepts and help identify issues that warrant further investigation.

Those are all legitimate uses. However, AI also has recognised limitations. It may draw incorrect conclusions from the facts provided, overlook an exception in the legislation, misinterpret Inland Revenue guidance, fill gaps in incomplete information with unsupported assumptions, or generate convincing but inaccurate references or case citations, or confidently present an answer that is simply wrong. Like any research tool, the quality of its output depends on both the information it is given and the judgement applied by the person using it.

That means the relevant question is not whether AI was used, but how it was used.

Reasonable care still requires judgement

Inland Revenue's Interpretation Statement IS 26/04, Shortfall penalty for not taking reasonable care, does not specifically address AI. Nor does it suggest that the use of technology changes the test for reasonable care.

Instead, the Interpretation Statement reinforces the long-established principle that reasonable care is assessed objectively by considering what a reasonable taxpayer would have done in the same circumstances.

The level of care expected will depend on the circumstances. Straightforward tax matters may require relatively limited enquiry. More complex transactions or positions with significant tax consequences generally require greater investigation and verification before a position is adopted.

Those principles apply equally where AI has been used. Using AI to assist with research or drafting is not, in itself, inconsistent with taking reasonable care. However, applying those principles in an AI context suggests that relying on AI-generated output without appropriate review may make it more difficult to demonstrate that reasonable care has been taken.

A useful way to think about AI is to treat it like a junior member of the team. It can produce a helpful first draft, identify relevant issues and improve efficiency, but its work should still be reviewed before it forms the basis of advice or a tax position.

AI cannot strengthen a weak tax position

The same reasoning applies when considering Inland Revenue's Interpretation Statement IS 26/05, Shortfall penalty for taking an unacceptable tax position.

Whether a tax position amounts to an unacceptable tax position depends on whether, based on the underlying legal analysis, it is about as likely as not to be correct, not on whether that analysis was prepared by a person or generated by AI.

If AI overlooks a relevant statutory provision, misunderstands the facts or produces legal analysis that results in a tax position that is not about as likely as not to be correct, the fact that sophisticated technology generated the answer does not improve the legal strength of the tax position. The position must stand on its own merits.

In other words, AI is simply another tool used in developing the analysis. It is not a substitute for sound legal reasoning.

Using AI well

Used appropriately, AI can improve efficiency, reduce administrative work and help practitioners focus on more complex advisory issues.

The challenge is ensuring that efficiency does not come at the expense of professional judgement. Some sensible practices include using AI to prepare research summaries or first drafts rather than final conclusions, checking AI-generated analysis against legislation, relevant case law and Inland Revenue guidance, verifying any AI-generated citations against the original source documents, ensuring the AI has been provided with complete and accurate facts, and exercising independent professional judgement before relying on any output.

These are not new expectations. They are simply the same standards that have always applied whenever practitioners rely on information produced by another source.

The takeaway

AI has become another valuable tool in the tax practitioner's toolkit. It can make tax work faster and, in many cases, more efficient. What it cannot do is change the standard expected of taxpayers or the professional judgement expected of advisers.

The important question is therefore not whether AI contributed to the advice. It is whether the taxpayer or adviser applied the level of care that would reasonably be expected in the circumstances before relying on that advice.  In that respect, nothing has changed.