Court Denies Motion to Prohibit Use of GenAI by Opponents

A court declined to ban AI in filings, so verifying the output stays with whoever signs it

In Boyd v. Beachwood City School Dist. Bd. of Education, 2026 WL 3181031 (N.D. Oh.), decided in October 2026, the defendants asked for an order barring the plaintiffs from using generative AI in any filing. As a fallback, they asked for mandatory disclosure of AI use. They relied on the federal sanctions rule and the court's inherent authority. Their lead example was a motion for leave to file a second amended complaint that cited a school board policy that does not exist. The court refused the ban. It found the hallucinated citations were sanctionable violations but imposed no sanctions "at this time". EDRM carried the report on October 29, in a post by Michael D. Berman. I have worked from that account, not the opinion.

Does this ruling mean courts will now tolerate AI in filings?

No. The court refused a blanket ban on one set of litigants and withheld sanctions for now. It also said plainly that the hallucinated material broke the sanctions rule.

The court held that because the plaintiffs are subject to the sanctions rule, that rule is an adequate safeguard. It did not need a tool-specific prohibition. It acknowledged AI's benefits and its risks. It also said the bar on citing hallucinations applies equally to pro se parties, while calling AI a potentially "important tool" for access to the courts for people who cannot afford counsel.

Why did the court decline to sanction despite the hallucinations?

The plaintiffs admitted the invented policy and accepted responsibility. The fabricated case law appeared in a motion irrelevant to the pending matters, so the court used its discretion against penalties.

The plaintiffs had argued their error was factual, about the contents of a policy, and not invented cases. The defendants' reply disputed that. It listed fabricated case law, misrepresented holdings and fictitious quotations, and said this undermined the plaintiffs' assurances that they had checked their work. The court noted that, outside the TRO motion, the defendants had not alleged hallucinated case law in any other paper.

That tally is the defendants' own count, and the court did not adopt it. The "at this time" wording is also a warning. A repeat in a motion that matters could go differently.

Is "it looks like AI" evidence of anything?

Not much on its own. The defendants cited formatting habits: vertical lists, heavy boldface, stray code tags. The court's concern was fabricated content, which can be checked against the record.

Style is a poor proxy for provenance. Plenty of human drafters over-bold and over-list, and the defendants' real evidence was a policy that could not be found. The same logic applies to your own matters. A rule that asks "was AI used?" is hard to police. A rule that asks "does this cite something real?" is not.

One gap: EDRM's account does not say how the court handled the alternative request for disclosure. Read the opinion before assuming it settled that point.

What should I ask my own vendor?

Ask whether every AI-generated statement links to a source document a human can open in one step. Ask also how the system records who verified it.

This case concerned litigants' filings, not review platforms, so it is not direct precedent for e-discovery tooling. The principle carries over, though. The signer owns the output, and courts will not excuse a fabrication because software produced it. A cited answer that opens the underlying exhibit turns verification into a click. A generic chat tool makes it a trip to the library.

Press vendors on their accuracy claims too. Ask who defined "hallucination rate", what test set was used, and whether the vendor or an independent party ran the measurement. Ask what the system does when a citation cannot be resolved. Does it fail closed, or does it show you a plausible answer anyway?

Frequently asked questions

Was anyone sanctioned?

No. The court exercised its discretion not to impose sanctions in this instance and not to restrict future AI use.

Does this protect lawyers who use AI?

Not in any broad way. The court's reasoning leaned on the plaintiffs' acknowledgement of their obligations. Counsel carry the same sanctions exposure for what they sign.

Is this binding elsewhere?

It is a single district-court order, reported by a commentator. Treat it as an example of judicial temperament, not settled law.

Sources: EDRM, "Court Denies Motion to Prohibit Use of GenAI by Opponents" (Michael D. Berman).