The IRS's Electronic Tax Administration Advisory Committee Recommends a Public Dashboard Explaining How the IRS Uses AI

The Electronic Tax Administration Advisory Committee, ETAAC, made its annual report to Congress in June 2026, published by the IRS as Publication 3415. Its members, in the words of the report's opening message, are "tax practitioners, enrolled agents, state and local administrators, payroll providers, and tax software experts." The report's third recommendation is that "The IRS should modernize its technology stack and approve the use of artificial intelligence with clear, documented transparency to preserve public trust." Parts of that recommendation follow, not the whole report.

Approving AI Use With Documented Transparency

"In May 2024, the IRS issued comprehensive guidance through Internal Revenue Manual (IRM) Section 10.24.1, and the Department of the Treasury maintained a public inventory of AI use cases that included the IRS." Then, in an interim policy of March 11, 2025, the IRS suspended that section of its manual pending further guidance from the Office of Management and Budget and the Treasury, and the Treasury's inventory of AI use cases, the report says, "is no longer publicly available."

As of the spring of 2026, it says, the federal AI policy landscape "continues to evolve." Whatever the instruments become, ETAAC holds that two principles should be kept and strengthened. The IRS should continue to approve and deploy AI "where it can be shown to deliver clear public benefits, consistent with applicable law and policy." And that approval "should be paired with proactive, plain-language transparency about how AI is being used, supported by documented governance."

A Dashboard in Plain Language

ETAAC recommends that the IRS publish, and regularly update, "a single AI disclosure dashboard or webpage written in plain language." For each current use of AI, it should show how the process was handled before AI, what tasks the AI is conditioned to perform, how the process is handled after AI was put in place, and how the IRS is mitigating risk where the use is rights- or safety-impacting. It should also say whether and how data is shared with third parties, and the data security measures in place.

Its list asks, at minimum, for appropriate information about expert review that validates that the IRS applies accepted best practices to mitigate bias where it uses standard anonymization or de-identification methods, for example to use taxpayer data to train or fine-tune generative AI models, "recognizing that public disclosure of specific enforcement methodology could compromise enforcement effectiveness." Where AI is used in a sensitive process such as audit selection or fraud detection, it says the IRS "should still disclose as much detailed, declassifiable information as possible," including the categories of inputs used, the role of human reviewers, and the channels available for taxpayers to challenge or seek explanation of an AI-assisted action.

Errors, Employees and Trust

On problems with AI tools, ETAAC says business units that deploy AI projects "should have an easy, well-publicized way to report technical issues or inaccuracies, and should be both incentivized and obligated to report them," to the project's technical owner for immediate remediation and to the IRS's project management office for AI, so that issues can be tracked for the annual review contemplated by IRM 10.24.1.5.3. It says anonymous employee surveys, protected against any potential repercussions, can give the IRS a candid view of how employees use AI, and that regular internal user surveys on in-house AI tools, combined with continuous post-launch testing, will help the IRS decide which tools to keep and which to retire.

On trust, the report says, "Mistrust in AI generally stems from a lack of understanding, which sensationalized misinformation readily exploits. The most effective protection against that dynamic is for the IRS to put a complete, accurate, and easy-to-understand account of its AI use directly in front of the public."