AI Standing Orders in Ohio Federal Courts: Scope and Proof of Violations

Ohio federal judges have published standing orders restricting the use of AI in preparing court filings.

For litigators, these orders raise two practical questions: what work does each restriction order cover, and how can a violation be established? The first question matters early, because compliance may begin before a brief is drafted: a restriction on preparing filings may reach not only the final text but also the research and record summaries used to develop the brief.

Some orders prohibit AI use outright. In the Northern District of Ohio, for example, Judge Christopher A. Boyko’s one-paragraph standing order prohibits attorneys and self-represented parties from using AI “in the preparation of any filing submitted to the Court.” It exempts information gathered from legal and internet search engines, including Westlaw, LexisNexis, Google, and Bing. District Judge Charles Esque Fleming takes a substantially similar approach (and uses the same “in preparation of” language) in Section 8 of his standing order. Potential sanctions under these orders include striking a pleading, monetary sanctions, contempt, and dismissal.

The research exceptions in these orders are not uniform. The Boyko and Fleming orders permit information gathered from legal search engines, such as Westlaw and LexisNexis, and internet search engines, such as Google and Bing, but never define what “in the preparation of” means.  Magistrate Judge Jonathan D. Greenberg’s standing order, likewise bans generative AI in preparing filings and permits information from legal search engines, but it also expressly prohibits Google and Bing AI Overviews. A Google search that is permitted under the Boyko and Fleming orders may therefore produce an AI Overview that Greenberg’s order prohibits. Differences like this are only the start of the scope question.

The first question, what work a restriction covers, is harder than it appears. The orders that prohibit AI use address “the preparation of” filings, which raises questions beyond who wrote the final text. Does an order cover AI used to summarize testimony, suggest an argument, or proofread a draft? This leaves open whether using Claude or Harvey or other LLMs for research, analysis, or other non-drafting uses is covered by the order or not. The provisions do not address each of these uses separately, and they do not expressly exempt work that a lawyer later verifies. Counsel should therefore assess both the accuracy of the filing and whether the methods used in its “preparation” comply with the applicable order.

The second question, how a violation can be established, starts with the reporting requirements: when does suspected AI use become discovered AI use that must be reported? The orders discussed above require parties and counsel to notify the court immediately if they discover AI use in a document filed in the case. Greenberg’s order refers specifically to generative AI. These provisions do not specify how disputed allegations should be resolved or expressly require counsel to test opposing filings with detection software.

Without that guidance, counsel may be tempted to infer AI use from circumstantial signs, but a brief’s writing style or an incorrect citation does not, by itself, establish AI use. Detection software also has limits. For example, reports from the National Institute of Standards and Technology (“NIST”) on detecting AI-generated content describe the challenges posed by text that combines human and AI writing and how paraphrasing AI-generated text can hinder detection.[1]

Even if AI use is confirmed, the next question is whether that use violated the applicable order. That turns on what the tool produced, what was done with it, and what the order permits. Information gathered through a search engine may be permitted, while an argument drafted with an AI tool may not. The line can blur, because a permitted search may return AI-generated output, as with the AI Overviews discussed above. A detector’s score alone cannot show which of those occurred.

For litigation teams, the practical steps are to check the assigned judge’s requirements at the outset, communicate them to everyone contributing to filings, and understand how the supporting work was prepared. If a dispute arises, both questions come down to the same facts: what tool was used, how it was used, and whether that use violated the order.

Endnotes

[1] Nat’l Inst. of Standards & Tech., NIST AI 700-1, 2024 NIST GenAI (Pilot Study): Text-to-Text Evaluation Overview and Results 1–2, 42–43 (2025), https://doi.org/10.6028/NIST.AI.700-1; Nat’l Inst. of Standards & Tech., NIST AI 100-4, Reducing Risks Posed by Synthetic Content: An Overview of Technical Approaches to Digital Content Transparency 28, 33 (2024), https://doi.org/10.6028/NIST.AI.100-4.

* * * *

If you have questions, please reach out to the author(s).

The Commercial Litigation Update blog is currently edited by Jennifer O'Connor and Adam Paine.