People Are Using AI to Do Their Taxes. Nobody Is Checking the Work.

Millions of self-employed individuals and small business owners are turning to AI tools like ChatGPT and Claude to handle their tax computations, without professional oversight.

The problem is straightforward. General-purpose AI models lack reliable mechanisms for staying current with tax law changes. These systems were trained on datasets with cutoff dates and cannot automatically update when tax rates shift, thresholds change, or deductions are eliminated.

Critically, the AI presents its work confidently and professionally, making errors difficult for non-accountants to detect. The output is well-formatted, the terminology is correct, and the reasoning sounds coherent.

The scale is substantial. Tens of millions of individuals worldwide are filing returns based on AI-generated workings with no verification. Accountants typically only enter the picture after problems emerge: tax authority inquiries, suspected underpayment, or one-off reviews of work already completed.

What the profession can do

Accountants need practical responses. Updated engagement letters reflecting different review scopes for AI-prepared materials. Verification of current legislation against what the model claims is true.

More fundamentally, we need structured, jurisdiction-specific tax knowledge infrastructure maintained by qualified practitioners. This is the foundation for more reliable AI-assisted tax computation - not better prompts, but better underlying knowledge.

This is part of what I'm building with Open Accountants: jurisdiction-specific accounting rules written and maintained by qualified accountants, structured so that AI systems can use them reliably.

Errors made with quiet confidence now outpace those made with self-doubt. The profession needs to catch up.