Obi v. Cook County: $5,000 Sanction — 30+ AI Hallucinations Including Fabricating the Court’s Own Prior Order

Background: A Federal Challenge to State Court Proceedings and a Filing Record

Ifeoma Delliane Chinedu Obi, proceeding pro se, filed suit in the Northern District of Illinois challenging Cook County Circuit Court proceedings. Her complaint was dismissed on November 18, 2025. Twelve days later she filed a Rule 59(e) motion to alter or amend the judgment — a 10.5-page single-spaced filing (equivalent to 21 pages double-spaced) without a table of contents, in violation of Local Rule 7.1.

This was not her first AI-related problem. In a prior filing, defendants had identified at least 17 instances of fake cases, quotes, and statements of law and fact from AI hallucinations. The court had extended “grace” at that time in light of her pro se status. She had now exhausted that leniency.

The AI Issue: Fabricating the Court’s Own Prior Order

Chief Judge Kendall documented the most egregious collection of AI hallucinations in a single docket yet published. The Rule 59(e) motion alone contained 13 identified hallucinations. The problems included:

• Fabricated quotation from Marshall v. Marshall, 547 U.S. 293 (2006): Plaintiff quoted the Supreme Court for the proposition that “Federal courts may exercise jurisdiction even when the case touches upon matters traditionally reserved to state law, where federal rights are at issue.” That quoted language is not in Marshall.
• Nonexistent Seventh Circuit case: Plaintiff cited Andrade v. Arby Concessions, 88 F.4th 1014 (7th Cir. 2023). “Andrade” is not a real case.
• Fabricated text from the court’s own prior order: Plaintiff asserted that “Page 7 [of the Court’s November 18, 2025 Order] explicitly references that ‘a federal court may hear a claim alleging a state court judgment is void for lack of jurisdiction.'” That language does not appear in the court’s order. The AI hallucinated text attributed to the sitting judge’s own words.
• Reply brief fabricated quotation from Ridder v. City of Springfield: Plaintiff quoted the Sixth Circuit as stating “Sanctions under Rule 11 are unavailable unless served before the case is disposed of.” That language does not appear in Ridder.

The court’s characterisation was unsparing: “Plaintiff’s motion is so replete with fabricated law, quotes, and statements that there is virtually no cogent argument actually supported by real law.”

Holdings

1. Motion struck for Local Rule violations. The oversized, table-less brief violated Local Rule 7.1 and was struck on those grounds independently of the AI issues.
2. $5,000 Rule 11 sanction. The court cited multiple recent Northern and Southern District of Indiana cases as comparators: $1,500 for two hallucinated cases; $6,000 for multiple hallucinated briefs; $7,500 for hallucinated citations; and a recommended $10,000 for a brief with hallucinated cases. The $5,000 sanction reflected the severity of the ongoing, repeated pattern.
3. “Grace” exhausted. The court explicitly noted it had extended leniency for the prior 17-hallucination filing but would not do so again. This two-strike framework has become standard: courts will forgive once but not twice.
4. Fabricating court order text is the most serious variant. The AI’s fabrication of text attributed to the court’s own prior order — quoting language back to the judge that the judge never wrote — is the most audacious and legally dangerous form of AI hallucination documented in published case law.

“Plaintiff’s motion is riddled with AI hallucinations, made up cases, quotes, and statements of law and fact… Plaintiff’s motion is so replete with fabricated law, quotes, and statements that there is virtually no cogent argument actually supported by real law… Plaintiff has exhausted that leniency.”

— Chief Judge Virginia M. Kendall, N.D. Illinois, April 9, 2026

India Angle: Fabricating Judicial Language — The Most Dangerous AI Variant

The most alarming element of Obi v. Cook County — fabricating text from the court’s own prior order — has a direct Indian parallel. Indian advocates sometimes quote from the court’s own previous orders in proceedings (“As Your Lordship observed on…”). If an AI tool generates a plausible-sounding judicial observation attributed to the sitting judge’s prior order, and that observation does not appear in the actual order, the advocate has put words in the judge’s mouth. This is not merely an AI citation error — it is, potentially, a contempt of court.

Relevant Indian Law

• Contempt of Courts Act 1971, Section 2(b) and (c): Wilful disobedience of a court order is civil contempt; scandalising the court is criminal contempt. Attributing fabricated language to a court order — even if done through AI rather than intent — could constitute criminal contempt if the court finds the advocate “ought to have known” the quoted language was not in the order.
• Advocates Act 1961 Section 35 / BCI Rules: Fabricating judicial language is the most severe form of misleading conduct prohibited under BCI Rule 9. The Bar Council disciplinary process would treat this as a serious professional misconduct warranting suspension.
• Code of Civil Procedure Order XLVII (Review): If an AI-generated fabricated quotation attributed to a court’s prior order influenced a subsequent order, the affected party would have grounds to seek review or clarification. But the advocate who submitted the fabricated quotation would face consequences separate from the review.

Three Practical Tips for Indian Practitioners

1. Never rely on AI to recall or summarise what a court said in a prior order in the same case. When drafting submissions that reference a court’s prior order, open the actual order document and quote directly from it. AI tools have no access to your case’s specific docket and will generate plausible-sounding but fabricated judicial language if asked to summarise.
2. One warning from a court is one warning — treat it as a strict prohibition. The “grace exhausted” framework in Obi shows that courts give one grace period for AI hallucinations, then enforce strictly. In India, where judicial warnings can be express (written) or implied (strong oral observations), treat any court comment about your citation accuracy as a zero-tolerance threshold going forward.
3. Volume is not a substitute for quality in any proceeding. The Obi brief was 21 pages double-spaced without a table of contents. Oversized, poorly organised AI-generated filings attract adverse attention regardless of their citation accuracy. Indian courts regularly limit written submissions in length; excessive volume is its own red flag.

Quick Takeaways

• 30+ AI hallucinations across two filings — the largest documented collection in published case law — resulted in a $5,000 sanction, the highest for a pro se litigant.
• Fabricating text attributed to the sitting court’s own prior order is the most dangerous AI hallucination variant — it directly impersonates the judge and fundamentally undermines the integrity of the proceedings.
• The “one grace period” framework is now standard: courts will extend leniency for a first AI hallucination episode, but subsequent violations are met with sanctions without further accommodation.
• Local rule violations (oversized brief, no table of contents) compound AI hallucination issues — compliance with procedural rules is not optional just because AI makes large filings easy to generate.
• In India: fabricating a court’s own judicial language is the most serious form of AI-related misconduct, potentially engaging both the Contempt Act and BCI disciplinary jurisdiction simultaneously.

Deep Dive: The Obi Sanction Scale and Where $5,000 Fits

The Obi order provides a rare judicial survey of the AI sanction scale, citing four prior cases as comparators. From lowest to highest: $1,500 for two fabricated cases; $5,000 for 30+ hallucinations across two filings (this case); $6,000 for multiple hallucinated briefs; $7,500 for hallucinated citations; and a recommended $10,000 for a brief with hallucinated cases. This range suggests that AI hallucination sanctions in the $1,500–$10,000 range are becoming normalised in U.S. federal district courts.

For Indian practitioners, the question of what equivalent sanction scale would look like under Indian law is significant. The Contempt Act does not specify a monetary cap for civil contempt fines (though Rs. 2,000 is a traditional cap in some contexts, higher courts have awarded substantially more). More practically, Rule 57.07 (Ontario’s personal costs rule) has Indian parallels in Order XX of the CPC — courts can award costs personally against an advocate who engages in misconduct. The quantum in Indian cases varies widely but a pattern of AI hallucinations severe enough to warrant a $5,000 equivalent in U.S. courts would likely attract both a cost order and a Bar Council referral in India.

The fabrication of court order language — unique to Obi in the published record — represents the logical extreme of AI hallucination: when the AI has insufficient factual grounding (because it cannot access the actual case docket), it generates plausible-sounding content that attributes false statements to the very authority it is supposed to be addressing. This failure mode is inherent to AI tools that do not have access to real-time case documents. The lesson is simple: for any claim about what a specific court said in a specific case or order, quote from the original document.