AI-invented case citations: what courts are doing, and what records lawyers should keep

Last checked 12 November 2026. General information, not legal advice or professional-conduct guidance. TimeBinder does not check citations or detect AI output, and a dated record is not proof that a filing is accurate.

What courts have been doing

Reports from the last few weeks, from legal publications:

• Kleyman v. Kaloidis (New York). According to The Expert Record, a lawyer filed 19 submissions citing 23 fictitious decisions that he testified came from Claude and other generative AI tools. The court imposed the $10,000 maximum sanction, awarded $36,511.35 in fees and costs (about 80 hours of work tracing the false authorities) and referred him to a grievance committee. He said he did not know the tools could invent authority; he also said that after the problem was raised he used the same AI platforms to “verify” the citations.
• Flycatcher v. Affable Ave. (S.D.N.Y.). The National Law Review reports that Judge Failla struck a defendant’s submissions and entered default judgment in February 2026 after counsel filed briefs with fake citations across three filings; 14 of 60 cases cited in one brief were found not to exist. In October 2026 she awarded the reporting attorney $38,877.59 in fees and costs.
• Other 2026 matters. Minnesota Lawyer (9 November 2026) lists $110,000 in fines and fees against two Oregon attorneys, an indefinite Nebraska suspension where 57 of 63 cited authorities were faulty or nonexistent, and a large firm acknowledging a bankruptcy-court filing with nonexistent citations. Law360 reported on 10 November that a federal judge in Massachusetts refused to reverse sanctions that included fees and removal from a case.
• The other side’s fakes. The ABA Journal reports that judges in some cases refused fees or relief to counsel who did not point out fake citations in opponents’ filings, including a California Court of Appeal decision, and a Minneapolis court called it “troubling” that a lawyer had not cite-checked the opposing brief.

These are reports of individual outcomes, drawn from secondary sources. Read the orders before relying on any of them.

What actually prevents it

1. Read every authority you cite in a primary or trusted citator source, not an AI summary. The Flycatcher court held that submitting cases without reading them would breach Rule 11 even on the lawyer’s own account.
2. Never use the same AI tool to verify what it wrote. The New York example above shows why.
3. Check quotations and pin cites, not just case names. Reports describe invented quotes alongside invented cases.
4. Cite-check opponents’ filings too and raise problems promptly, in line with the court’s rules and your professional obligations.
5. Follow your court’s AI-disclosure or certification rules, and your firm’s policy.

Records worth keeping

None of this replaces the checking. But when a filing is challenged, an organised record shows what you did:

• A verification log: for each authority, the date you read it, the source used, the person who checked, and any pin cite or quotation verified.
• Dated drafts: the version filed, and earlier versions if drafting involved AI tools or junior staff. Draft history can show who changed what, and how an error entered. Our piece on working drafts covers the idea for writers.
• Filed and served copies, as received: keep the file exactly as served by the other side, so a later dispute about which version had the fake citation can be settled by comparing hashes. Docket copies help too.
• Your correction trail: the notice to the court, the date, and the corrected filing.

A SHA-256 fingerprint of each file (try our hash checker; the file stays on your device) and a TimeBind of the set records that exact files existed by a date. For exhibits and chain of custody, see our guide for lawyers and chain-of-custody log.

What a TimeBind does and does not show here

A TimeBind is a dated, checkable record that an exact file existed no later than a recorded time and still matches. Your first Pooled TimeBind is free; see how it works. Here it can show that your verification log and filed drafts existed in a given form by a given date.

It does not check citations, show that a filing is accurate, or show you acted reasonably. A timestamp on a bad brief is a true record of a bad brief. It is a supporting layer for honest process records, not a defence against sanctions.

Limits

This is general information, not legal advice or professional-conduct guidance, and it relies on secondary reports. Court rules on AI use and disclosure differ by court and judge. Check the rules for your court and speak to your regulator or ethics counsel where unsure.

Tools that can help

There is no affiliate link on this page: nothing here depends on a paid third-party tool. Our affiliate disclosure explains when we do use them.

Sources checked

• The Expert Record, Kleyman v. Kaloidis: Claude Citations and Sanctions. Read via search extract; read the order itself before citing.
• National Law Review, New York Judge Finds Counsel Misused AI Across Three Filings (Flycatcher Corp. v. Affable Ave. LLC, S.D.N.Y.). Read via search extract.
• Minnesota Lawyer, It’s no hallucination: AI-generated errors persist in court filings, 9 November 2026. Read via search extract.
• Law360 Pulse, Atty Can’t Duck Sanctions Over Alleged AI Fake Citations, 10 November 2026 (headline and opening only; paywalled).
• ABA Journal, Judges are increasingly faulting lawyers who fail to point out AI-generated fake citations in their opponents’ filings. Read via search extract.
• Public interest on X (22 and 20 October 2026): two posts on catching the other side’s fake citations reached about 2,900 and 1,700 impressions; a Volokh post on 7 November reached about 600. Demand on X is modest; press coverage is steady.

Keep a dated record of your drafts and checks

Your first Pooled TimeBind is free, with no credit card required. After that, TimeBinds start from US$5, and an ASAP TimeBind is US$15. Click here to start, or see plans and pricing.