Our Take
Courts are now enforcing consequences for AI-generated hallucinations as if they were professional negligence, not tool failure—which means your firm's AI use is your liability, not the vendor's.
Why it matters
Legal professionals have treated generative AI as a research shortcut without audit controls. This ruling signals courts will hold practitioners accountable for factual errors embedded in AI output, making AI governance a liability issue, not an efficiency gain.
Do this week
Legal teams: audit all AI-assisted filings and briefs submitted in the past 18 months for citation accuracy before opposing counsel finds hallucinated citations.
Appeals court imposes sanctions for AI hallucinations in briefs
A US appeals court has sanctioned lawyers for submitting court filings that cited nonexistent case law generated by ChatGPT. The court found the lawyers failed to verify the AI-generated citations and did not disclose their use of the generative model to opposing counsel or the court. The sanctions were imposed on the basis of "lack of candor," a procedural violation distinct from incompetence.
The briefs in question contained fabricated case citations that appeared authentic but did not exist in any legal database. When the opposing party challenged the citations, the lawyers acknowledged they had used ChatGPT to research the cases without independently verifying the results.
Liability now rests with the user, not the tool
This ruling establishes a precedent that professional responsibility attaches to the lawyer, not to OpenAI or the AI vendor. Courts are treating AI hallucinations as a known failure mode that practitioners have a duty to catch before submission. The "lack of candor" framing is particularly significant: the court did not excuse the error as a reasonable reliance on a new tool. Instead, it treated undisclosed AI use and unverified output as a failure to meet professional standards.
For legal departments and solo practitioners, this means AI-assisted work is now subject to the same verification burden as any other research method. The court's decision signals that courts will not recognize AI hallucination as a mitigating factor in professional misconduct cases. Sanctions can include attorney fees, fines, and reputational damage.
What to do now
If you have submitted AI-generated content in any legal filing, disclosure memo, or formal brief in the past year, audit it immediately. Cross-reference every citation, statistic, and factual claim against primary sources before the opposing party does. Document your verification process; if sanctions are imposed, proof of post-hoc audit may reduce penalties.
Second, establish a written AI governance policy that specifies which tasks permit AI assistance (brainstorming, summarization, template generation) and which do not (legal research, fact-checking, citation generation). Require sign-off from a supervising attorney before any AI-assisted work enters a filing or client communication.
Third, disclose AI use to clients and courts when it has been part of your work product. Transparency on tool use is now a defense against sanctions; silence is not.