The IRS sent out a bulletin yesterday and we’re going to write about it because for once it’s not about tedious tax stuff that our audience doesn’t have the patience to slog through: Introductory Guidelines for Responsible AI Use in Federal Tax Practice. This is important stuff, people! You should probably read the whole thing if you work in this space. It’s not that long, promise.
It’s a fairly decent document full of warnings and specific Circular 230 citations, like this one:
- § 10.35 – Competence: “A practitioner must possess the necessary competence to engage in practice before the Internal Revenue Service. Competent practice requires the appropriate level of knowledge, skill, thoroughness, and preparation necessary for the matter for which the practitioner is engaged.”
Practitioners must understand both the law and the technology used in their representation of clients before the IRS, including AI systems’ operational mechanics, limitations, and risks.
They must understand how AI develops content, recognize the potential for bias or errors, and be able to evaluate whether AI outputs are suitable for use in IRS matters. Lack of technological competence could lead to improper advice or flawed filings.
In other words, they’re telling you straight up that “I didn’t think the AI would lie” is not an appropriate defense should AI lead you astray.
But let’s talk about this section. You’ll notice they use GAI to mean “generative artificial intelligence,” an acronym we really hope doesn’t catch on.
Real-World Consequences of Improper AI Use
Courts have increasingly sanctioned lawyers for improper use of GAI, mainly due to fake citations or other hallucinations contained in legal filings. Typical penalties imposed in these cases have included financial sanctions—often amounting to several thousand dollars; public censure; required completion of legal ethics or professional responsibility courses; default judgments entered against the responsible party; removal from representation of a party to the matter; and disciplinary referrals to state bar authorities.
These sanctions stem from violations of duties of candor and competence, and they often entail reputational harm as attorneys must notify clients and affected judges about the court-ordered sanctions. The consequences emphasize the need for tax practitioners to act with care and precision when using AI tools, reinforcing the importance of diligent human oversight.
While the legal profession has seen the most public court sanctions, they serve as a warning for other tax professionals. Cases imposing sanctions against other types of tax professionals are slowly emerging. For example, the Australian government published a 230-plus page report on its website in July 2025 that Deloitte Australia had prepared for it. The report contained invented quotes attributable to a judge, references to non-existent reports, and books ascribed to the wrong author, all produced apparently by GAI. Deloitte Australia reportedly resolved the matter directly with the Australian government by agreeing to partially refund a portion of the fee it received from the government.
“A partial refund looks like a partial apology for substandard work,” said Labor senator Deborah O’Neill to AFR in October. “Deloitte has a human intelligence problem,” she said, sharpening her tongue on Deloitte’s incompetence. “This would be laughable if it wasn’t so lamentable … too often, as our parliamentary inquiries have shown, these consulting firms win contracts by promising their expertise, and then when the deal is signed, they give you whatever [staff] costs them the least.”
Anyway, go read the bulletin and then do some homework on how LLMs work if you need to.
Newly minted 
