SERVANTSTACK // INCIDENT INTELLIGENCEFORENSIC IMAGE // VERIFIED FRAMEThe Sanction Wave
In a Single Month, Courts in Michigan, New York, Ontario and Italy All Punish Lawyers for AI-Fabricated Citations - as the Global Case Count Hits 1,667
In under two weeks, four courts in three countries sanctioned lawyers for filing AI-hallucinated authority.
- 01TRIGGERIn under two weeks, four courts in three countries sanctioned lawyers for filing AI-hallucinated authority.
- 02MACHINE ACTIONAdvisory output
- 03MISSING GATERisk-based SME approval before execution
- 04IMPACTRights & due process
The short version
In under two weeks, four courts in three countries sanctioned lawyers for filing AI-hallucinated authority.
Case telemetry
- INCIDENT
- SS-IR-091
- DATE
- June 11-23, 2026
- SYSTEM
- The Sanction Wave
- LOCATION / SCOPE
- United States, Canada, Italy
- EVIDENCE
- Official finding
- AI ROLE
- Advisory output
- HARM
- Rights & due process
- SOURCES
- 2 cited records
The event
In under two weeks, four courts in three countries sanctioned lawyers for filing AI-hallucinated authority. On June 11, Italy's Corte di Cassazione fined counsel EUR 5,000 plus costs for hallucinated precedents in a criminal appeal. On June 12, the Law Society Tribunal of Ontario ordered CAD 31,150 in full-indemnity costs against a lawyer who used Grok and filed fabricated citations. On June 17, the Michigan Court of Appeals, in Barber v. Morawa, held an attorney personally liable for the opponent's damages and fees after his briefs - and even his "Notice of Correction" - contained fresh AI-fabricated cases, and referred him to the Attorney Grievance Commission. On June 23, New York's Appellate Division ordered $10,500 in sanctions in Landberg v. City of New York over fabricated cases and fictitious Court of Appeals quotations. The AI Hallucination Cases database logged 81 such court decisions in June 2026 alone, bringing the verified worldwide total to 1,667. Even the elite tier is not immune: in April, Sullivan & Cromwell apologized to a federal bankruptcy judge after an emergency motion in In re Prince Global Holdings contained AI hallucinations the firm's review protocols failed to catch.
What the machine did
General-purpose chatbots - ChatGPT, Grok and their peers - generate legal authority the way they generate everything else: fluently, confidently, and without any connection to whether the case exists. Each lawyer treated the model's output as research rather than as unverified text. The Michigan case shows how deep the failure runs: when opposing counsel flagged the fake citations, the attorney asked the tools for a fix and filed a "correction" that itself contained new fabrications. At Sullivan & Cromwell, the firm admitted its review protocols existed but were not followed - the checkpoint was on paper, not in practice.
Where the failure landed
Monetary sanctions on two continents, personal liability for an opponent's legal fees, referrals to bar disciplinary bodies, and a growing body of published precedent holding that citing AI output without verification violates the duty of reasonable inquiry. The volume is the story: 81 decisions in one month means courts have moved from novelty warnings to routine enforcement, and every filing signed by a human who did not read what the machine wrote is now a documented professional hazard - for solo practitioners and white-shoe firms alike.
Official finding
Supported by a court, regulator, inquiry, or other official record cited below.
SOURCE RECORD UPDATED 2026-07-09
2 cited records
- 01
- 02
Risk-based SME approval before execution
The failure pattern in this case: High-stakes output had no accountable checkpoint.
The moment the path could change
The appropriate subject-matter expert reviews the evidence, exceptions, and affected people before the output becomes action.
Autonomy is a design choice.
See the operating model that keeps AI useful while preserving human authority at consequential moments.
Compare AgenticAI and AugmentedAI →