A text-scanning tool flagged more than 50 federal appeals court opinions from 2026 as showing signs of AI authorship, with one opinion scoring over 50% likely AI-written.
Quick Take
- A researcher tested about 2,250 published federal appeals court opinions from January through early August 2026 using an AI-detection tool called Pangram.
- More than 50 opinions showed signs of AI authorship, with scores ranging from under 1% to over 50%.
- A baseline test of over 300 opinions from January 2022, before generative AI tools were widely used, found zero signs of AI writing in any of them.
- The finding follows a string of cases where judges and lawyers were caught using AI that produced fake case citations or fabricated quotes.
- Judicial ethics guidance already warns judges they may use AI for research but should never let it decide a case’s outcome.
Text Scan Flags Dozens of Opinions
A writer working with Reason magazine’s Volokh Conspiracy blog ran a large-scale check on federal appeals court writing. Using an AI-detection tool called Pangram, the analysis covered nearly every published opinion from the regional circuit courts between January and early August 2026, about 2,250 documents total.
More than 50 of those opinions showed signs of AI authorship. The scores varied widely, from less than 1% likely AI-written up to more than 50% in a few cases. Most of the flagged opinions clustered near the low end of that range, suggesting light AI involvement rather than full AI-drafted rulings.
Old Opinions Show a Clean Baseline
To check the tool’s reliability, the researcher also scanned more than 300 published circuit opinions from January 2022, years before generative AI tools became common in legal writing. Every single one came back clean. Pangram found no AI signals in any passage or sentence, with each opinion scoring exactly 0.000000% AI-written. That stark contrast between the two time periods gives the 2026 findings real weight.
Judges Have Already Owned Up to AI Use
This isn’t the first sign that AI has crept into judicial writing. In 2025, two federal judges withdrew opinions after acknowledging they included generative AI that produced fake, “hallucinated” case citations. Judge Julien Xavier Neals of New Jersey and Judge Henry T. Wingate of Mississippi both pulled their rulings and apologized. Senator Chuck Grassley later released judges’ written responses admitting to AI use, and called for continued oversight of the practice.
Rules Say AI Can Help, Not Decide
Legal ethics guidance already addresses this exact problem. Groups that write rules for judges say a judge may use AI for research, but “should NEVER use AI to reach a conclusion on the outcome of a case.” Any use of AI in drafting an opinion or order should be handled “with extreme caution”. Courts have also disciplined lawyers for filing briefs full of AI-invented case citations, with one California court ordering a $10,000 sanction over fabricated quotes.
Courts Are Already Debating AI’s Limits
Federal courts have started weighing in directly on how much AI should shape legal decisions. In one case, the Fifth Circuit commented after a district judge described in a newspaper interview how he used artificial intelligence to help with his decision-making. Separately, the D.C. Circuit ruled that AI cannot legally be considered an “author” at all, rejecting a copyright claim for a work generated entirely by a computer program.
Together, these developments point to a judiciary still figuring out where AI fits. The tool-based text analysis doesn’t prove any single judge broke the rules. But it lines up with a growing paper trail of judges admitting AI involvement, bar associations issuing warnings, and courts sanctioning lawyers for AI mistakes. For a system built on human judgment and public trust, that combination raises real questions about how opinions affecting real people’s lives get written.
Sources:
reason.com, mayerbrown.com, ncji.org
