ATL280

ATL280: Proposed AICPA-IRS Tax AI Risk Framework

• Episode: ATL280
• Podcast: Accounting Technology Lab
• Hosts: Randy Johnston and Brian F. Tankersley, CPA.CITP, CGMA
• Recorded: October 2, 2026
• Published: October 23, 2026
• Duration: Approximately 25 minutes
• Presented by: CPA Practice Advisor

Episode summary

Artificial intelligence is changing tax preparation, but who is responsible when the technology gets it wrong? In Episode 280 of the Accounting Technology Lab, Randy Johnston and Brian Tankersley examine a proposed AI tax risk framework discussed in materials associated with former IRS Commissioner Danny Werfel. They focus on five categories of risk for tax preparers: technical, practice, legal and regulatory, business, and workforce and strategic. The hosts examine AI hallucinations, inaccurate document extraction, client data leakage, overtrust of AI-generated tax advice, and erosion of professional expertise. They also explain how AI-driven productivity can overwhelm conventional human review procedures. Other concerns include shadow AI, professional liability, client disclosure, vendor dependence, and preservation of institutional knowledge. The discussion distinguishes risks faced by practitioners from risks facing tax authorities, identifying hallucinations, overtrust, skill erosion, data leakage, and vendor dependence as overlapping concerns. The practical message is that faster preparation does not eliminate professional accountability. Firms need independent verification, appropriate human oversight, reliable source documentation, protection of confidential taxpayer information, and thoughtful AI governance. The framework is discussed as a proposal, not as a binding new IRS rule.

Key takeaways

• Verify AI tax conclusions against authoritative sources.
• Preserve meaningful human review even as AI throughput increases.
• Protect taxpayer information from unauthorized tools and vendors.
• Clarify professional liability, disclosure, and governance responsibilities.
• Maintain tax expertise and contingency plans for vendor dependence.

Quotable moments

Time	Speaker	Transcript excerpt
05:55	Brian Tankersley	“The hope here is ... that they’re going to keep their weeds level knowledge of this and not just outsource the whole thing to AI.”
07:38	Brian Tankersley	“Just imagine a TikTok tax advice.”
11:29	Randy Johnston	“The AI answers are being gained by people who have certain messages they want to appear.”
12:34	Brian Tankersley	“The scary part about it, though, is that many people don’t go back and look at primary sources.”
14:20	Randy Johnston	“More returns get processed with less scrutiny.”
19:09	Brian Tankersley	“The AI is supremely confident all the time.”
24:00	Randy Johnston	“I think it’s a must-read for anybody that’s in practice.”

Social media posts

1. LinkedIn

AI is changing tax preparation. But when AI gets the tax law wrong, who is responsible? ATL280 explores the proposed AI tax risk framework. #TaxAI #ATL280

2. LinkedIn

Skill erosion may be the AI risk firms underestimate. Can your staff catch an AI-generated tax mistake? #TaxProfessionals #ATL280

3. LinkedIn

AI agents can prepare work in parallel. Human review is still a bottleneck. How should firms respond? #AccountingAI #ATL280

4. LinkedIn

One AI error can threaten years of client trust. Learn why disclosure and review matter. #TaxPractice #ATL280

5. LinkedIn

A confident AI answer is not authoritative tax guidance. Verify the primary sources. #TaxResearch #ATL280

6. X

AI can prepare a return. Can it defend the tax position? #TaxAI #ATL280

7. X

AI can be confidently wrong. Tax professionals need independent verification. #ATL280

8. X

More returns. Faster processing. Less review? That is not automatically progress. #ATL280

9. X

Shadow AI and confidential tax data are a dangerous combination. #DataPrivacy #ATL280

10. X

Your AI vendor may change. Your professional responsibility does not. #ATL280

11. Facebook

AI makes tax preparation faster, but does it make it safer? ATL280 explores five categories of risk.

12. Facebook

Are we losing the tax expertise needed to catch AI errors? Randy and Brian discuss skill erosion in ATL280.

13. Facebook

Would you trust a tax answer just because it sounds confident? Explore AI hallucinations and overtrust in ATL280.

14. Facebook

What happens when one AI tax error affects hundreds of clients? ATL280 examines professional liability and reputational risk.

15. Facebook

Your employees may already use unapproved AI tools. ATL280 discusses shadow AI and client data.

16. Instagram

AI is fast. Tax law is complicated. Professional judgment matters. #TaxAI #ATL280

17. Instagram

Five AI risk categories. One question: Who checks the AI? #ATL280

18. Instagram

Hallucinations. Overtrust. Data leakage. Skill erosion. Vendor dependence. #ATL280

19. Instagram

AI sounds confident. Your tax research still needs authoritative sources. #TaxResearch #ATL280

20. Instagram

Do not let AI turn your review process into a rubber stamp. #AccountingAI #ATL280

Editorial and sourcing notes

• Based on the October 23, 2026 episode transcript and timestamped SRT.
• Framework discussed as proposed; this page does not claim it is a binding IRS regulation.
• Timestamp labels are approximate segment start times.