AI in Tax Preparation: Balancing Innovation and Accountability - SmarterArticles S1E12

#podcast #ai #ethics #tax

Written by Tim Green, narrated by AI. Listen to the full episode here.

🎙️ Season 1, Episode 12 | Duration: 15:58

Mike Todasco's viral experiment filing taxes with OpenAI Codex made headlines in 2026. The promise was seductive: upload your documents, let an AI agent handle the complexity, and get a better refund. But behind the marketing, the reality is far more concerning.

This episode examines the accountability gap at the heart of AI tax tools: taxpayers bear all the legal risk while AI providers bear none. Along the way, it covers reliability failures in leading models, Intuit's own admission that generative AI "performs poorly at math," and a 2026 court ruling that strips legal privilege from AI-generated work.

This episode uses AI voice narration from ElevenLabs Studio.

The Reliability Gap

Testing by the New York Times and the TaxCalcBench benchmark reveals that leading AI models often miss small details and fail strict field-by-field accuracy checks. Even advanced multi-agent systems fall short of full automation. The errors are not edge cases: they include misclassifying income, omitting deductions, and producing calculations that would trigger audit flags on a real return.

Intuit, for its part, has acknowledged that generative AI "performs poorly at math" and avoids using it for TurboTax calculations. When one of the industry's biggest players publicly distances its core product from the technology it markets, the disconnect between promise and performance becomes hard to ignore.

Who Bears the Risk

Taxpayers remain legally responsible for every error on their return. The AI tool that made the mistake faces no professional liability, operates under no professional standards, and is not required to disclose its limitations. This creates a one-sided risk: you bear the consequences, the AI provider bears none.

The Privilege Problem

A 2026 ruling in U.S. v. Heppner denied privilege for AI-generated legal work. The implications for tax preparation are significant. Information you share with a human tax professional is protected by attorney-client or tax-preparer privilege. Information you share with an AI tax tool may not be. If that information becomes discoverable in proceedings, the taxpayer who relied on the AI tool has lost a core legal protection through no fault of their own.

The Regulatory Divide

The IRS publicly warns against relying on AI for tax preparation, yet uses AI extensively in its own enforcement operations. Meanwhile, the EU AI Act classifies tax-related AI as high-risk, imposing a compliance framework covering transparency, accuracy, and human oversight. The gap between these approaches leaves taxpayers in very different positions depending on where they live.

What Would Help

Three things would begin to close the accountability gap:

Professional liability for AI tax tools: the same standards that apply to human preparers should apply to AI providers
Mandatory disclosure of accuracy rates: taxpayers deserve to know how often these tools get it wrong
Clear privilege protections: taxpayers should not lose legal protections simply because they used an AI tool instead of a human professional

Until those safeguards exist, the advice from the IRS, ironically, is worth following: if you use AI for taxes, you are on the hook for everything it gets wrong.