When AI Files Your Taxes: Who Pays When It Fails #
Tax season 2026 arrived with a peculiar new ritual. Across kitchen tables and home offices, millions of filers uploaded W-2s, 1099s, and brokerage statements not to a human accountant, but to an algorithmic system promising speed, savings, and superior accuracy. The pitch was irresistible: why pay thousands for a professional when an AI agent can ingest your financial life, cross-reference the tax code, and spit out an optimised return in minutes?
One early adopter, Mike Todasco, documented the experiment on his Substack in vivid detail. He pointed OpenAI's Codex at a folder of tax documents, fed it a master prompt, and waited. Three hours and roughly twenty dollars later, the system had processed his return, a task that would have cost him around ten thousand dollars with his usual accountant. The post went viral. The implication was unmistakable: the AI tax revolution had arrived, and it was cheap.
But here is the question nobody racing to upload their documents seems to be asking. When the algorithm gets it wrong, and the evidence suggests it will, who exactly picks up the bill?
The Allure of the Algorithmic Accountant #
The shift from tax software to tax agents is one of the defining themes of the 2026 filing season. Having AI “do” your taxes now means deploying large language models and agentic AI systems that pull data from financial institutions, read blurry 1099-K photographs using optical character recognition, categorise thousands of Venmo transactions, reconcile brokerage statements, and surface recent changes in tax law. Intuit, the company behind TurboTax, has gone all in on what it calls “done-for-you” experiences. Its AI engine, Intuit Assist, uses both traditional and generative AI to provide personalised recommendations, flag potential errors in real time, and even deploy a specialised agent, the “1099 Cost Agent,” that can ingest supplemental PDF forms and reason through stock sales to identify the correct cost basis.
Intuit announced in early 2026 that it had paired advanced agentic AI with a nationwide network of 13,000 human experts, creating what it describes as the only all-in-one consumer platform for year-round personal finance management. Credit Karma's Tax Assistant, another Intuit product, claims that members with simple tax situations who answer quick questions throughout the year can have up to 80 per cent of their Tax Year 2025 returns ready to go by filing time. TurboTax Live Assisted is marketed as “the only tax filing solution on the market that provides customers an expert final review at no added cost, ensuring 100 percent accuracy and maximum refund guaranteed.” That guarantee, notably, applies to the human-reviewed product, not to the AI outputs alone.
The competition is just as aggressive. H&R Block launched AI Tax Assist, a product designed to streamline preparation for individuals, the self-employed, and small-business owners. Newer entrants like Hive Tax AI can pull in years of past financial data, automatically organise transactions, and help identify missed deductions. TaxGPT markets itself as an AI tax assistant for individuals, promising to simplify the filing process through conversational interfaces. The message from every corner of the industry is the same: the machines are ready.
Yet the machines, it turns out, are not nearly as ready as the marketing suggests.
When the Maths Does Not Add Up #
In early 2025, The New York Times conducted a test that should give every aspiring AI tax filer . Reporters ran eight fictional tax scenarios, developed in partnership with tax-filing service TaxSlayer, through four leading AI chatbots: Google's Gemini, OpenAI's ChatGPT, Anthropic's Claude, and xAI's Grok. The chatbots were provided with all necessary forms. The result was sobering. On average, the tools miscalculated the refund or amount owed to the IRS by more than two thousand dollars.
The Times attributed the failures to a fundamental design limitation: AI chatbots do not truly understand the complex relationships among the pieces of information they process, and errors accumulate as tasks become more interconnected. Benedict Evans, a prominent technology analyst, told the newspaper that “the problem with taxes is all those very small little details matter, and it's not going to get every single little detail right.” He acknowledged that the models improve dramatically every six months, but added that they still only give “roughly the right answer,” which is not sufficient for taxes.
The nature of these failures matters as much as their frequency. Large language models are probabilistic systems. They generate outputs based on statistical patterns in their training data, not by executing deterministic calculations. This means that the same input can produce different outputs on different runs, a characteristic that is fundamentally incompatible with the precision required in tax preparation. As multiple experts have noted, the results are “unexplainable” in the formal sense: you cannot go back and audit the reasoning chain the way you can with traditional tax software, where every calculation is traceable to a specific rule in the code.
Independent benchmarking has confirmed the scale of the problem. TaxCalcBench, a rigorous evaluation framework created by Column Tax and published on arXiv in July 2025, tested frontier models on their ability to calculate personal income tax returns. The benchmark uses 51 test cases representing a range of personal tax situations, and a return is considered “correct” only if every evaluated field matches the expected value exactly, reflecting the IRS's own standard. The results were stark. Gemini 2.5 Pro, the best-performing standalone model, achieved just 32.4 per cent strict accuracy. Claude Opus 4 managed 27.5 per cent. GPT-5 reached 41.7 per cent. Common failure modes included consistent misuse of tax tables, errors in tax calculation, and incorrect eligibility determinations.
Even Filed, a company using a multi-agent architecture with validation layers, only achieved 72.5 per cent strict accuracy on complete federal returns, though it reached 94 per cent on a line-by-line basis. Patrick McKenzie, the well-known fintech commentator, has cited 2026 to 2028 as the AI industry's consensus window for when large language models might genuinely be able to “do taxes.” Column Tax itself concluded that the task is likely not automated by the end of 2026, and that achieving it will require strong tax domain expertise and proprietary datasets that go well beyond what general-purpose language models currently possess.
NerdWallet published its own analysis in March 2026, testing ChatGPT, Gemini, and Perplexity on seven tax questions. The team combed through more than 50,000 words of chat transcripts and found that while the chatbots performed well on black-and-white questions, they produced inconsistent answers when the same question was asked multiple times and made assumptions about users that could lead to personalised errors. Sam Taube, NerdWallet's lead writer for investing and taxes, noted that “a couple of years ago, even the cutting-edge AI models couldn't reliably do basic arithmetic,” and that while recent updates have improved their maths skills, “the tendency to cite nonexistent, 'hallucinated' cases in response to legal questions still comes up in 2026.” His summary was blunt: “Taxes involve both of those subjects, math and law. It's not a reliable source of truth yet.”
There is an uncomfortable irony here. Intuit's own vice president of product management has publicly acknowledged that generative AI “doesn't do well with math yet,” which is why TurboTax does not use AI for its actual calculations. Making sure tax code outcomes are accurate, the executive said, is “always job number 1A,” adding: “We don't feel that generative AI is at a point yet where it can do that.” The company that sells the most popular tax software in the world is telling you, in effect, that AI cannot do the thing that millions of people are increasingly using AI to do.
The Accountability Void #
If the accuracy picture is complicated, the liability picture is worse. When you sign your tax return, you attest under penalty of perjury that the information is accurate to the best of your knowledge. The IRS holds you accountable for your return's accuracy regardless of what tools or methods you used in preparation. There is no special category for AI-assisted errors. No safe harbour protects you from liability based on reliance on algorithmic outputs. If the AI is wrong, the IRS treats that error as your mistake. This creates a structural asymmetry that ought to trouble anyone who has uploaded a PDF to a chatbot and clicked “file.” The companies building these tools bear minimal liability for the advice they generate. No contract exists between you and the AI in any meaningful sense. No professional liability insurance covers AI errors. No licensing board can sanction an algorithm for providing incorrect advice. The terms of service for virtually every consumer AI product disclaim responsibility for the accuracy of outputs, often in language buried deep in documents that almost nobody reads.
The contrast with traditional tax preparation is instructive. When you hire a human accountant or a CPA, that professional is bound by licensing requirements, ethical codes, and professional liability standards. If they make an error, there are established mechanisms for recourse: malpractice claims, professional disciplinary proceedings, and often errors-and-omissions insurance that can cover the financial damage. None of these mechanisms exist for AI tax tools. The technology occupies a regulatory gap between “software tool,” which carries product liability, and “professional service,” which carries professional liability. It is treated as neither, and thus escapes both frameworks.
Laura Carrubba, an accounting instructor at George Mason University, has warned bluntly that filers should “never, ever upload any kind of sensitive personal information into a public forum like that.” The privacy risks alone are substantial, but the liability exposure is arguably worse. As one tax professional put it to reporters: “The alibi can't be that ChatGPT told me to do it; that's kind of equivalent to the dog ate my homework.”
For tax professionals who use AI tools in their practice, the picture is somewhat different but no less fraught. Practitioners remain professionally liable for supervising AI-generated advice, ensuring its accuracy in the context of intricate tax laws and client-specific circumstances, and validating recommendations before presenting them to clients. AI developers may bear some responsibility for tool reliability, but current service agreements shift most liability to users. As one widely cited legal analysis put it, “the blame game is perhaps the same as it ever was; the responsibility for competent advice lies with the tax professionals who employ these and other tools.” Canadian tax professionals have already reported a troubling pattern. A survey found that businesses are losing money after relying on AI tools for financial and tax advice, with tax professionals spotting mistakes on a regular basis. The problem, they warn, is not hypothetical. It is materialising now.
A Landmark Ruling and Its Ripple Effects #
The legal landscape shifted significantly in February 2026, when Judge Jed Rakoff of the Southern District of New York issued what appears to be the first ruling to squarely address privilege claims involving generative AI. In United States v. Heppner, the defendant, a corporate executive charged with securities fraud, wire fraud, and making false statements to auditors in connection with an alleged scheme to defraud investors of approximately 150 million dollars, had used a consumer version of Anthropic's Claude to research legal issues related to the government's investigation.
Without his lawyers' direction, Heppner inputted information he had learned from his attorneys into the AI platform, generating roughly thirty-one documents that outlined defence strategy and potential arguments. Federal agents seized these documents during the search of his residence after his arrest in November 2025.
Judge Rakoff ruled that the AI-generated documents were not protected by either attorney-client privilege or the work product doctrine. His reasoning was direct. Claude “is not an attorney,” and the platform's privacy policy specified that it collects data on user inputs and outputs, uses that data to train the tool, and reserves the right to disclose such data to third parties, including governmental regulatory authorities. There was no confidentiality. There was no legal advice. There was no privilege.
The decision, described by the court as addressing “a question of first impression nationwide,” sent shockwaves through the legal and financial services communities. The New York State Bar Association published an analysis under the headline “Loose AI Prompts Sink Ships,” underscoring the severity of the implications. The Harvard Law Review noted that the conclusion was not as inevitable as Judge Rakoff's opinion might suggest, arguing that a more fact-intensive analysis would indicate that self-directed AI use should be privileged in at least some circumstances. But the practical implications are already reverberating through corporate tax departments, law firms, and compliance teams. The ruling raises pressing questions for any organisation incorporating AI into its workflows: if an employee feeds sensitive client data into a consumer AI tool to generate tax analysis, is that analysis discoverable? The answer, after Heppner, appears to be yes.
Judge Rakoff left open one important possibility. He suggested that the analysis might differ if AI use had been directed by counsel under a Kovel-type arrangement, where the AI could “arguably be said to have functioned in a manner akin to a highly trained professional who may act as a lawyer's agent within the protection of the attorney-client privilege.” This distinction between supervised and unsupervised AI use may prove to be one of the most consequential legal questions of the coming years.
The Regulatory Vacuum #
The IRS itself has taken notice of AI's incursion into tax preparation, though its response so far has been more cautionary than prescriptive. For the first time in history, the agency addressed AI on its annual Dirty Dozen list of tax scams for 2026, warning about AI-enabled IRS impersonation via phone calls, AI-generated phishing content, and voice cloning. Nina Tross, liaison for tax advocacy at the National Society of Tax Professionals, told reporters that “AI is definitely the number one culprit” for perpetrating tax scams. Bad actors, she explained, use AI to gather information from taxpayers and corporations, then file “highly detailed” fraudulent tax forms that result in improper payments.
The IRS has also explicitly cautioned against relying on AI for tax guidance, reminding taxpayers that they “should not rely on AI-generated responses to complex tax questions” and should verify any calculations or information provided by artificial intelligence. But the agency has stopped well short of issuing comprehensive standards for AI use in tax preparation.
This regulatory gap is drawing increasing criticism. Bloomberg Law has reported on growing calls for federal leadership, noting that accounting software companies are promoting AI-powered tools to taxpayers while sidestepping responsibility for errors and passing liability to clients. A letter sent to Treasury Secretary Scott Bessent urged comprehensive federal guidance on AI use in tax preparation, warning that without it, a patchwork of conflicting state rules would undermine business compliance and CPA professionalism. The comparison to the employee retention credit scheme, which earned its place on the IRS's own Dirty Dozen list, is apt: unregulated AI in tax preparation threatens to become the next entry.
Meanwhile, the IRS itself is quietly embracing the technology internally. The agency now operates 129 AI use cases, up from 54 in 2024, with AI powering audit selection, fraud detection, and taxpayer services. Yet the IRS has provided minimal public information about how its algorithms work, and taxpayers selected for audit are not told whether it was humans or AI that flagged their return. The asymmetry is striking: the government uses AI to scrutinise your return, but disclaims responsibility when you use AI to prepare it.
Across the Atlantic, the European Union's AI Act offers a more structured approach. The legislation, which entered into force on 1 August 2024, classifies AI systems by risk level and imposes corresponding obligations. Many AI use cases common in financial services, including credit scoring, fraud detection, and automated decision-making that affects access to services, are explicitly classified as high-risk, subject to strict requirements around risk management, human oversight, transparency, and auditability. For tax advisory firms specifically, the AI Act requires that operators ensure employees possess adequate AI literacy, that chatbots be clearly recognisable as AI systems, and that client data not be entered into open generative AI models without anonymisation. The European Banking Authority published a factsheet in November 2025 on the AI Act's implications for the banking and payments sector, and in November 2025 the European Parliament adopted a resolution laying out its priorities for AI use in financial services.
The full obligations for high-risk systems were initially set to take effect on 2 August 2026, though the European Commission proposed in November 2025 to extend that deadline to December 2027. FINRA in the United States expects compliance frameworks to be operational by the fourth quarter of 2026, with examinations beginning in early 2027.
A peer-reviewed study published in Nature's Humanities and Social Sciences Communications in 2025 examined how AI-driven systems impact legal fairness, due process, and the integrity of tax procedures. The researchers identified risks including algorithmic bias, opacity, and weakened procedural safeguards, and proposed an independent AI oversight mechanism to explain and review tax decisions. The study's central argument is that without such mechanisms, the use of AI in tax administration risks undermining the very principles of fairness and transparency that tax systems are built upon.
The Profession Fights Back, and Adapts #
The accounting profession's response to the AI incursion has been a mixture of anxiety and strategic repositioning. A recent survey found that over half of financial services professionals, some 52 per cent, believe their job prospects have worsened in the past year due to AI, while 57 per cent avoid raising concerns with managers due to job insecurity. The World Economic Forum's Future of Jobs 2025 report listed accountants, auditors, and bookkeepers among “the world's fastest-declining jobs,” predicting 92 million global job displacements by 2030, with AI cited as a primary driver. Studies from OpenAI and the International Labour Organisation have also identified accountants and tax preparers as occupations “highly exposed to disruption.”
Yet the profession simultaneously faces a severe talent crisis. More than 300,000 accountants have left the profession since 2020, and three-quarters of CPAs are approaching retirement age. Recruitment agency Robert Half observed growing demand for accountants in 2025, with 58 per cent of employers planning to increase their permanent finance and accounting headcount, a six-percentage-point rise from 2024. The Bureau of Labor Statistics projects 5 per cent growth in accounting through 2034, with 124,200 annual openings. Surveys show that 46 per cent of firms intend to hire more full-time staff and 45 per cent plan to hire more seasonal staff, even as more than a third anticipate automating processes using AI.
The resolution to this apparent paradox lies in the profession's deliberate pivot from routine compliance work toward advisory services. Routine bookkeeping faces an estimated 85 per cent automation risk, but advisory roles face under 25 per cent. Tax professionals are shifting from two-hundred-dollar return preparation to planning engagements worth five to twenty-five thousand dollars, handling multi-entity structures, international tax planning, audit representation, and strategic advice that demands human judgement and client trust.
The American Institute of CPAs launched its Profession Ready Initiative on 2 February 2026, a research-backed effort to identify and develop the skills early-career CPAs need in an AI-driven marketplace. Susan Coffey, CEO of public accounting for the AICPA, described the initiative as addressing “one of the accounting profession's most pressing needs.” The research, led by SkillEdge, a firm specialising in professional practice analysis, will examine the roles early-career CPAs perform, how job expectations align against education curricula, and where professionals need additional development support. The organisation is developing a framework around the “T-shaped professional,” combining deep expertise with broad capabilities in analytics, digital fluency, and strategic thinking.
New roles are already emerging. Firms are hiring AI compliance officers to ensure ethical and audit-ready AI use, exceptions managers to handle discrepancies that AI cannot resolve, and AI audit reviewers to oversee investigations as auditing moves from sampling to full-visibility analysis. Notably, one of the Big Four accounting firms has already announced plans for an end-to-end AI audit process in 2026. CPA Practice Advisor published a pointed essay in February 2026 warning that if the profession lets software do all the thinking, firms risk becoming “interchangeable,” because if every CPA provides the same computer-generated answers, clients will simply pick the cheapest option. The industry's emerging consensus is captured in a phrase that has become something of a mantra: “AI handles the 'what.' A great accountant tells you 'so what' and 'now what.'”
The Trust Deficit #
Consumer sentiment tells a more complicated story than the breathless headlines about AI tax filing might suggest. A YouGov study released in January 2026 found that just 19 per cent of Americans trust AI in financial services, and only 10 per cent trust AI to make financial decisions automatically. Yet the 2026 IPX1031 Tax Procrastinators Report found that 46 per cent of Americans say they trust AI for tax advice, while 21 per cent said they would use AI to help them actually prepare their returns this year.
The gap between these figures hints at something important. People may tell pollsters they trust AI for tax advice, but far fewer are willing to hand over full decision-making authority. This is the uncanny valley of financial automation: close enough to useful to be tempting, far enough from reliable to be dangerous. The distinction between using AI as an assistant and using it as a replacement is one that the marketing rarely makes clear, but it is the distinction upon which financial safety depends.
Early IRS data for the 2026 filing season shows more than 36.5 million refunds totalling roughly 136.6 billion dollars issued as of early March, with the average refund running approximately 10.6 per cent higher than at the same point in 2025. Part of this increase may reflect the complexity of the One Big Beautiful Bill Act, the sweeping federal tax package passed in July 2025 that reshaped parts of the US tax code with new credits and deductions. This is precisely the kind of legislative complexity that trips up AI systems. This year's return is not simply last year's return with minor adjustments; it is a substantially different document, and the models trained on prior-year data may not have fully absorbed the changes.
Asking Harder Questions #
The convenience narrative around AI tax filing is seductive, and not entirely wrong. For a straightforward W-2 return with no complications, an AI assistant may well produce an adequate result, particularly when integrated into established tax software that uses deterministic calculation engines for the actual maths. The problems begin at the margins, and in taxation, the margins are where the money is.
Consider the filer with cryptocurrency holdings across multiple exchanges, or the freelancer juggling 1099 income from several states, or the small business owner navigating the new provisions of the One Big Beautiful Bill Act. These are precisely the scenarios where AI chatbots have been shown to fail most spectacularly, and they are also the scenarios where the financial consequences of an error are most severe. An incorrectly claimed deduction does not just cost you the deduction itself; it can trigger an audit, generate penalties and interest, and in extreme cases, result in criminal liability for making false statements on a federal return.
The deeper issue is not whether AI will eventually get good enough at taxes. It almost certainly will. The issue is what happens in the interim, while millions of filers are being encouraged to trust systems that independent benchmarks show cannot correctly calculate even a third of federal returns. The consumer protection framework for this transition period is essentially nonexistent. There is no required disclosure when an AI system generates tax advice. There is no mandatory accuracy threshold. There is no insurance requirement. There is no regulatory body specifically overseeing AI tax preparation tools.
What would a responsible accountability framework look like? At minimum, it would require transparency about when AI is generating tax advice versus when a deterministic engine is performing calculations. It would mandate accuracy benchmarks, perhaps modelled on TaxCalcBench, that AI tax tools must meet before being marketed to consumers. It would require some form of liability insurance or indemnification, so that taxpayers who rely on AI advice in good faith are not left entirely on their own when the algorithm gets it wrong. And it would establish clear regulatory oversight, whether through the IRS, the Federal Trade Commission, or a new body entirely, to ensure that the gap between marketing claims and actual capability does not continue to widen.
This is the accountability gap that demands urgent attention. The technology is advancing faster than the legal and regulatory frameworks designed to govern it. Companies are marketing AI tax tools with confidence-inspiring language while their own engineers acknowledge the technology is not ready for the task. Taxpayers are absorbing all the risk while the companies building these tools absorb none of it.
The question is not whether we should celebrate the convenience. Convenience is fine. The question is whether we are willing to build the accountability structures that make that convenience safe, before the next filing season, and the one after that, and the one after that, turn millions of taxpayers into unwitting participants in the largest unregulated experiment in financial automation the world has ever seen.
The IRS will not accept “the AI did it” as an excuse. Perhaps it is time we stopped accepting it from the companies selling these tools, too.
References and Sources #
- Todasco, M. “Yes, I Did My $10,000 Taxes With a $20 AI.” Substack, 2026.
- The New York Times. AI chatbot tax accuracy test using eight fictional tax scenarios with ChatGPT, Gemini, Claude, and Grok, 2025.
- Intuit Inc. “Intuit's AI-Driven Expert Platform Redefines Tax Filing with 'Done-For-You' Experiences.” Intuit Investor Relations, 2026.
- Intuit Inc. “Intuit's All-in-One Agentic AI-Driven Consumer Platform Powers Year-Round Money Outcomes.” Intuit Investor Relations, 2026.
- Column Tax. “TaxCalcBench: Evaluating Frontier Models on the Tax Calculation Task.” arXiv, July 2025.
- Filed. “Measuring AI Tax Accuracy: Comparing Filed to ChatGPT, Claude, and Gemini on an Open Benchmark.” Filed.com, 2025.
- NerdWallet. “Analysis: What AI Gets Right (and Very Wrong) About Taxes.” NerdWallet.com, 3 March 2026.
- Morgan Lewis. “Using AI in Tax Workflows? What Heppner Means for Tax Departments.” MorganLewis.com, March 2026.
- Harvard Law Review. “United States v. Heppner.” Harvard Law Review Blog, March 2026.
- New York State Bar Association. “Loose AI Prompts Sink Ships: How Heppner Shook the Legal Community.” NYSBA.org, 2026.
- Internal Revenue Service. “Dirty Dozen Tax Scams for 2026.” IRS.gov, March 2026.
- Bloomberg Law. “IRS Standards on AI and Tax Preparation Would Protect Businesses.” Bloomberg Law, 2026.
- Nature Humanities and Social Sciences Communications. “Balancing Innovation and Integrity: AI in Tax Administration and Taxpayer Rights.” Nature.com, 2025.
- European Commission. “AI Act: Shaping Europe's Digital Future.” Digital-strategy.ec.europa.eu, 2024-2026.
- Cross Border Advisory Solutions. “EU AI Regulation in Tax Law: New Obligations for Tax Advisory Firms.” CrossBorderAdvisorySolutions.com, 2026.
- Accounting Today. “Accounting and Tax Staff Worry AI Threatens Jobs.” AccountingToday.com, 2025.
- World Economic Forum. “Future of Jobs 2025 Report.” WEForum.org, 2025.
- AICPA. “AICPA Launches Profession Ready Initiative to Transform CPA Workforce Readiness.” AICPA-CIMA.com, 2 February 2026.
- CPA Practice Advisor. “The Decline of Human Intelligence in Tax Strategy: Is AI Replacing Smart Accountants?” CPAPracticeAdvisor.com, 16 February 2026.
- YouGov. AI in Financial Services Trust Survey. January 2026.
- IPX1031. “2026 Tax Procrastinators Report.” IPX1031.com, 2026.
- Robert Half. Accounting and Finance Hiring Survey. 2025.
- Bureau of Labor Statistics. Occupational Outlook Handbook: Accountants and Auditors. BLS.gov.
- Capitol Technology University. “Audited by an Algorithm: How the IRS Is Using AI in 2026.” Captechu.edu, 2026.
- OpenAI and International Labour Organisation. AI Occupational Exposure Studies. 2024-2025.
Tim Green UK-based Systems Theorist & Independent Technology Writer
Tim explores the intersections of artificial intelligence, decentralised cognition, and posthuman ethics. His work, published at smarterarticles.co.uk, challenges dominant narratives of technological progress while proposing interdisciplinary frameworks for collective intelligence and digital stewardship.
His writing has been featured on Ground News and shared by independent researchers across both academic and technological communities.
**ORCID:** [0009-0002-0156-9795](https://orcid.org/0009-0002-0156-9795)
**Email:** [tim@smarterarticles.co.uk](mailto:tim@smarterarticles.co.uk)
Listen to the free weekly [SmarterArticles Podcast](https://www.smarterarticles.fm)