When AI Gives Tax Advice: Who Is Responsible?
By AG CPA Co. | Perspectives on Finance & Technology | ~650 words
A taxpayer spends forty minutes with an AI-powered tax assistant. The system reviews their uploaded documents, identifies deductions, calculates their liability, and generates a filing recommendation. The taxpayer follows it. The return is filed. Six months later, an IRS notice arrives: the deduction was impermissible. The penalty is real. The question is deceptively simple: who is responsible?
This question is no longer hypothetical. AI-assisted tax guidance is already embedded in consumer software, professional platforms, and enterprise compliance systems. As the tools become more capable and more widely used, the accountability gap between what these systems can do and what legal frameworks currently require has become one of the more pressing issues in tax administration.
The Accountability Problem
Traditional tax advice comes with a clear chain of responsibility. A CPA or enrolled agent provides guidance, signs a return, and stands behind it — professionally, ethically, and legally. Circular 230, the Treasury Department's set of regulations governing practice before the IRS, establishes standards of conduct and creates mechanisms for discipline when those standards are violated. The practitioner is identifiable. The advice is attributable. The liability is assignable.
AI systems complicate each of these elements. A Tax Notes analysis of automated tax planning examined the question directly, concluding that existing frameworks are poorly equipped to assign liability when an AI system provides incorrect or incomplete guidance. The system itself cannot hold a Circular 230 designation. The software vendor may disclaim responsibility through terms of service. The tax professional who incorporated the tool may argue they relied on it reasonably. The taxpayer is left in a gap between all three.
A 2025 UK tribunal case illustrates how these tensions are already reaching courts. In Elsbury v Information Commissioner, the First-tier Tribunal ordered HMRC to disclose whether it had used generative AI in correspondence to taxpayers about research and development tax relief claims — the first time an English court confronted the question of algorithmic transparency in tax administration directly. The case surfaced a broader tension: tax authorities are using AI at scale while taxpayers and courts are demanding accountability for how those systems make decisions.
Certification and Oversight
One proposed solution is mandatory certification for AI systems used in tax contexts. Under this framework, vendors would be required to submit their systems for evaluation against defined standards before deploying them in advisory roles. Connecticut has already moved in this direction, directing its Department of Administrative Services to inventory and assess AI systems used by state agencies for potential discriminatory effects. Extending this logic to tax advisory AI would at minimum create a documented record of what the system was designed to do — and what it was not.
But certification has limits. Tax law changes continuously. A system certified in January may be producing incorrect guidance by April. Continuous reevaluation, as some have proposed, would require infrastructure that does not yet exist and raises its own question: who conducts the audits, and under what authority?
The Human in the Loop
A parallel approach focuses not on regulating the AI itself but on preserving the human professional's role in the chain of advice. Under this model, AI tools assist with research, data extraction, and preliminary analysis, but a licensed professional reviews, takes responsibility for, and signs the output. The Wolters Kluwer 2025 Future Ready Accountant Report found that 40 percent of North American accounting firms already use AI-driven tax research, with nearly 80 percent planning to increase that investment — suggesting this hybrid model is already the operational norm. The regulatory question is whether to formalize it.
The risk is that formalizing human review requirements creates a compliance checkbox rather than genuine oversight. If a professional is processing hundreds of AI-generated returns per day, meaningful review becomes difficult to sustain.
A Question Worth Taking Seriously
The current moment asks something important of practitioners, policymakers, and the public alike. AI systems in tax contexts are not neutral tools. They make recommendations that carry real financial and legal consequences. The frameworks that govern accountability — professional ethics standards, software liability law, tax administration regulations — were not designed with these systems in mind.
Getting this right matters not because AI tax tools are bad, but because they are increasingly consequential. The practitioner who understands this intersection — who can help clients navigate both what the technology can do and what the regulatory environment requires — is occupying territory that will only become more important.
REFERENCES
Tax Notes. 'Automated Tax Planning: Who's Liable When AI Gets It Wrong?' September 2023.
Elsbury v Information Commissioner [2025] UKFTT 915 (GRC). UK First-tier Tribunal, August 2025.
Wolters Kluwer. '2025 Future Ready Accountant Report.' 2025.
Fieldfisher. 'AI and Tax: Litigation, Risk, Use Cases.' October 2025.