A Reuters headline dated October 22, 2026 says a US appeals court lifted the block on Mississippi's ban on diversity, equity and inclusion programs and ejected the district judge from the case over AI errors. We could not open the full Reuters story or the appellate opinion, so the final outcome below is attributed to that headline and should be treated as reported, not independently confirmed. What we can document in detail is how a judge's order ended up with fabricated content in the first place, and what the Fifth Circuit asked about it. The background is well sourced and is the part that matters most for anyone using AI in legal or other high-stakes work.
TL;DR: confirmed versus reported
| Question | Status |
|---|---|
| Did Judge Henry Wingate's July 2025 order contain AI-related errors? | Confirmed by the judge himself and by the ABA Journal |
| Which tool was used? | Perplexity, used by a law clerk, per Wingate's own letter |
| Was it a docketing mistake? | Wingate called it an early draft that was docketed by mistake |
| Did the Fifth Circuit press on the issue? | Yes, at argument on August 31, 2026 |
| Did the Fifth Circuit lift the block and remove the judge? | Reported by Reuters on October 22; we could not read the ruling |
| Is the law itself now in effect? | Not independently confirmed here; check the Fifth Circuit docket |
What happened to the original order
In July 2025, US District Judge Henry T. Wingate of the Southern District of Mississippi issued a temporary restraining order blocking a state law that bars diversity, equity and inclusion practices in public K-12 and higher education. The challenge was brought by the ACLU of Mississippi and the Mississippi Center for Justice.
According to the ABA Journal, the order referenced allegations and parties not in the lawsuit, nonexistent declarations by four people, and language not found in the state law being challenged. The Mississippi attorney general's office, defending the law, raised the problems, and Wingate withdrew the order and replaced it with a corrected version.
The Magnolia Tribune reported that Senate Judiciary Committee Chairman Chuck Grassley wrote to two federal judges in October 2025 about alleged use of generative AI to draft orders with little human verification, and that Wingate was one of them. In his reply to the Administrative Office of the Courts, Wingate said a law clerk used Perplexity "strictly as a foundational drafting assistant to synthesize publicly available information on the docket." He also wrote: "This was a mistake," adding that the opinion docketed on July 20, 2025 "was an early draft that had not gone through the standard review process."
He named the root cause as a lapse in human oversight: posting a draft instead of a final version and skipping final review. He said he addressed the clerk's conduct and put a new plan in place: every draft opinion gets a mandatory independent review by a second law clerk, and all cited cases are printed from Westlaw and attached to the final draft.
Timeline of the case
| Date | Event | Source |
|---|---|---|
| April 2025 | Mississippi governor signs the anti-DEI law for public schools and universities | Magnolia Tribune |
| June 2025 | ACLU of Mississippi and Mississippi Center for Justice sue | Magnolia Tribune |
| July 20, 2025 | Order with errors is docketed, per Wingate's letter | Magnolia Tribune |
| Two days later | The state, in consultation with plaintiffs, moves to clarify or correct | Magnolia Tribune |
| The next day | Wingate amends the order and removes the first from the docket; the outlet reports the amended order was backdated | Magnolia Tribune |
| October 2025 | Senator Grassley writes to two federal judges about AI-drafted orders | Magnolia Tribune |
| Late October 2025 | Wingate answers in a letter naming Perplexity and a draft that skipped review | Magnolia Tribune |
| August 31, 2026 | Fifth Circuit hears argument on reassignment | ABA Journal, Law.com |
| October 22, 2026 | Reuters headline reports the block lifted and the judge ejected | Reuters (not read in full) |
The backdating detail comes from a single outlet and we have not verified it against the docket, so read it as reported. It matters because the question of how a correction is handled, openly or quietly, is part of what makes AI mistakes in official records harder to trust.
What people are asking
Was the judge himself using AI? By his own account, a law clerk used it, and the judge took responsibility for the docketing failure. The legal system generally holds the judge accountable for what bears their signature.
Is this the same as a lawyer citing fake cases? Similar mechanism, different stakes. A sanctioned lawyer harms a client; an order with invented material shapes a state law's fate and the public record.
Does this mean judges cannot use AI? Nothing in the sources we read says so. Wingate's remedy was more review, not abandoning the tool. Policy bodies are still debating where to draw lines.
Does the substance of the DEI dispute depend on the AI error? The plaintiffs said it did not; the state said the injunction relied on the flawed reasoning. That disagreement is exactly what an appeals court has to settle.
What the Fifth Circuit asked
At argument on Monday, August 31, 2026, the Fifth Circuit weighed whether the case should go to a different district judge. The ABA Journal, citing Law.com, reports that Judge Jerry Smith asked why an erroneous citation still appeared in the amended order, saying: "Don't you think that's a pretty serious matter?"
Deputy Mississippi Solicitor General Anthony Shults argued the court should vacate the order and send the case to another judge, because the later preliminary injunction "embraced" and relied on the reasoning in the flawed restraining order. Robert McDuff of the Mississippi Center for Justice argued the injunction should stay because the judge, not an AI tool, made the decision. Other coverage reports Judge Kurt Engelhardt questioning how the appeals court could be confident the inaccurate material had not affected the district court's analysis.
The Reuters headline of October 22 suggests the panel sided with the state on both points: lifting the block and removing the judge. Confirm the exact holding, whether the injunction was vacated or reversed, and any instructions for remand, in the opinion itself.
Why this is an AI story and not only a legal one
The pattern is familiar to anyone who follows lawyers sanctioned over AI-hallucinated citations. The difference here is that the person responsible for catching errors was the decision maker, and the errors reached a published order that affected a state statute.
Three details matter for people who build or deploy AI tools:
- The failure was a workflow failure. Wingate's own words point to a draft that skipped review. A model that synthesizes docket material will sometimes invent a declaration or a quotation. The control that was missing was a human check before publication.
- A fix that is not applied everywhere does not fix it. The error that drew the Fifth Circuit's attention was one that survived into the corrected order, according to Judge Smith's question as reported.
- Reliance compounds. The state argued the later injunction leaned on the flawed reasoning. Once an AI-assisted document is treated as authoritative, downstream decisions inherit its mistakes.
This sits alongside other courtroom collisions with AI we have covered, such as the Arizona court that threw out a sentence over an AI video of the victim, and the vendor side of the debate in OpenAI's Astra for law, which pitches AI legal research as a professional tool. For the training angle, see how law schools are adapting in UChicago's AI-resilient legal education.
What to do if your team uses AI to draft anything that gets filed or published
The lesson transfers from courts to code review, compliance and research:
- Verify every citation against the source. Wingate's fix, printing each case from Westlaw and attaching it, is a concrete model. In software, the equivalent is running the tests and reading the diff. Our piece on AI-assisted pull requests and agent safety covers that.
- Name a human reviewer and make review a gate. Two sets of eyes with a documented sign-off beat a policy that says be careful.
- Separate drafts from finals in your system. Wingate described a draft that should never have been docketed. Use statuses, folders or branches so a draft cannot be published by accident.
- Disclose when asked. Several courts now require disclosure of AI use in filings; check local rules.
- Log what the tool saw. Wingate stressed that no sealed or confidential information was entered. Being able to say that matters.
If you deploy agents that act on documents or systems, AgentBeam, the agent security platform from the explainx.ai team, is built to stop agents before they take dangerous actions, which is a similar idea to a mandatory review gate applied to software.
What we do not know yet
- The exact wording of the Fifth Circuit's October 22 decision as reported by Reuters.
- Whether the panel made findings about sanctions or discipline, which this coverage does not address.
- Whether the plaintiffs will seek rehearing, which would delay the law taking effect.
- Whether Congress or the federal judiciary will act on AI rules for chambers; Grassley's inquiry in October 2025 asked about this, and the outcome is not covered in our sources.
We will update this post if the opinion or further reporting changes the picture. For broader policy context, the AI policy timeline for 2026 tracks the main regulatory moves.
Related reading
- Lawyers sanctioned over AI hallucinations
- Arizona court vacates sentence over AI victim video
- OpenAI Astra for law
- UChicago law and AI-resilient legal education
- AI-assisted pull requests and agent safety
- AI policy timeline 2026
Reported details are accurate as of October 23, 2026 per the sources linked above. The appellate outcome is attributed to a Reuters headline and had not been verified against the opinion.