Generative AI has made its way into the courtroom, and not always gracefully. A recent order from the U.S. District Court for the District of Colorado, Miller v. Colorado Department of Revenue, offers one of the clearer judicial statements on what happens when a filer relies on AI-generated authority without checking it. (Antonio Miller v. Colorado Department of Revenue, No. 25-CV-03167-NYW-TPO, 2026 WL 2931909 (D. Colo. Sept. 30, 2026)). The decision is worth attention not because the sanctions imposed are novel, but because of who got sanctioned, how the court identified the problem, and what it did in response.
The Setup: Colorado Court Flags AI-Generated Hallucinations in a Pro Se Filing
Antonio Miller, a servicemember proceeding pro se, sued the Colorado Department of Revenue over tax-collection actions tied to a restaurant he owned. After a magistrate judge recommended dismissing his claims, Miller filed objections. Before reaching the merits of his objections, Judge Nina Y. Wang paused to address something unusual about the authorities Miller cited: several of them “bear hallmarks of ‘hallucinated’ citations created by generative artificial intelligence.”
What follows is a catalog of takeaways for litigators.
Takeaway 1: AI Hallucinations Are No Longer Just a Lawyer Problem
The headline-grabbing “fake citation” sanctions cases have mostly involved attorneys who leaned on chatbots and filed the results unchecked. Miller extends the trend to self-represented litigants. The court was explicit that pro se status is not a shield: a litigant “still has a responsibility to ensure that the authority he relies on stands for the proposition that he seeks to cite the authority for.” The practical point is that AI misuse is spreading across every category of filer. It appears that courts are not going to grade hallucinated authority on a curve because the person who submitted it lacks a law license.
Takeaway 2: Hallucinations Come in Gradations
A particularly useful feature of the opinion is that it does not treat “AI hallucination” as a single phenomenon. The court identified several distinct defects in Miller’s citations: some contained “foundational errors, like citing to the wrong court,” while in other instances the filer provided “the correct case-identifying information but mischaracterize[d] the substance of the case.” In one example, a cited holding was described as the opposite of its actual outcome. That is the subtle danger: a hallucination is not always an invented reporter volume that a quick search exposes. It might be a real case, correctly cited, standing for a proposition it never announced. Those errors survive a cursory check and only surface when someone reads the opinion.
Takeaway 3: Courts Frame AI Misuse as a Threat to Case Administration
Judge Wang did not treat the problem as a mere inaccuracy. She located the harm in the basic machinery of adjudication, observing that “the misuse of generative artificial intelligence inhibits a court from adjudicating matters in a just, speedy, and economical manner,” and holding that the misrepresentations were sanctionable. That framing matters. The injury is not limited to the opposing party who must rebut a phantom case; it extends to the court’s docket and to public confidence in the system. We may see this rationale repeated as more courts confront the issue. (see also Oct. 9, 2026, decision of SDNY Judge Failla addressed at eDiscovery Watch here).
Takeaway 4: Inherent Authority Is the Enforcement Engine — and It Acts Fast
Miller’s opponent did not have to file a Rule 11 motion to trigger consequences. Exercising its inherent power, the court sua sponte struck the objections that contained misstatements and inaccurate citations and refused to grant leave to refile corrected versions. The lesson is that the fallout may be immediate and merits-affecting, not merely a monetary penalty down the road. A filing built on hallucinated authority may simply disappear from the record, taking the arguments with it.
Takeaway 5: Standing Orders and AI Certifications Have Teeth
The decision rests in part on the District of Colorado’s standing order, effective Dec. 1, 2025, requiring an AI-use certification on filings — which Miller’s objections lacked. The court then ordered Miller to show cause why he should not be sanctioned under Rule 11 and specifically directed him “to address whether he used any sort of artificial intelligence tools in drafting his objections and explain his efforts to verify the accuracy of the citations.”
For those practicing in multiple jurisdictions, this is the actionable compliance point: local AI rules are proliferating, and courts are enforcing them. Practitioners should know whether their court requires a certification and treat it as a substantive obligation rather than boilerplate.
Postscript: The Sanction Is Not Yet Baked
One caveat: Miller is a show-cause order, not a final sanctions ruling. The court directed Miller to respond by Oct. 21, 2026, so any sanction has not yet issued. GT is monitoring developments in this matter.