Andrew Blair-Stanek (Maryland), Nils Holzenberger (Institut Polytechnique de Paris) & Benjamin Van Durme (Johns Hopkins), ChatGPT Generates a Novel Tax Strategy, 189 Tax Notes Fed. 1097 (Nov. 17, 2025):

Three years ago in Tax Notes [Shelter Check: Proactively Finding Tax Minimization Strategies via AI, 177 Tax Notes Fed. 1515 (Dec. 12, 2022)], we predicted that artificial intelligence would be able to generate tax minimization strategies when fed only the text of tax authorities — without any internal IRS data and without any laborious hand coding of tax code sections. That day has arrived.

OpenAI’s industrial-grade reasoning large language model o1, which was under the hood of ChatGPT-o1, created what we believe to be the first publicly disclosed tax minimization strategy generated by a large language model. This came from our systematic exploration of how these models understand past tax strategies. …

Our systematic exploration of the capabilities of large language models produced what we believe to be the first novel tax strategy generated by these models. This strategy is fundamentally different from Notice 2003-54 in several ways. Although this strategy seems to pose little danger to the Treasury — it is probably a loser in court and likely a reportable transaction — it should be a wake-up call about the risks AI poses to tax revenues.

Editor’s Note:  If you would like to receive a daily email with links to tax posts on TaxProf Blog, email me here.