In June 2023 two New York lawyers were fined 5,000 dollars for filing six cases a chatbot had invented. It made global headlines. Three years on, a public database maintained by the researcher Damien Charlotin tracks roughly 1,490 court decisions worldwide in which a party relied on AI-fabricated material, more than a thousand of them in the United States, with new entries almost daily. The largest known penalty is now around 110,000 dollars, imposed by a federal court in Oregon. A federal appeals court has imposed 15,000 dollars per attorney. In April 2026 the Nebraska Supreme Court entered the first indefinite bar suspension tied to AI-fabricated filings. In June a judge in Mississippi cancelled a trial and suspended the lead lawyers on both sides from the district for two years after both had filed fake citations.
1. The tools are not the excuse
The early cases involved general-purpose chatbots. That defence is gone. In a Fifth Circuit matter this year counsel used two purpose-built legal research products and still filed sixteen fabricated quotations. Stanford's RegLab found that dedicated legal research tools hallucinate between 17 and 34 percent of the time, far less than chatbots and nowhere near zero. Grounded tools reduce the error rate; they do not remove the duty to read the opinion.
2. The cover-up is what gets punished
Across the decisions analysed by Norton Rose Fulbright and others, the aggravating factor is consistent. Fines follow fabrication; discipline follows denial. The Nebraska lawyer first denied using AI, then admitted it. The Fifth Circuit noted that greater candour would have produced a lesser penalty. Courts are explicit that admonitions have not worked and that they intend to move to punitive sanctions.
Candour with the court is the one thing the tool cannot supply for the person who used it.
3. Which rules are doing the work
In the United States most sanctions rest on Rule 11, which requires a reasonable inquiry into what you sign, and on the professional rules of competence, candour and supervision. Courts in the United Kingdom, Canada, Australia, Singapore and across the EU have converged on the same principle in the same period: verify before you file. For accountants, auditors and advisers the analogues are already in place in professional standards on due care and documentation. The legal cases are simply where the record is public.
4. What the firms that are not in the database do
- Verification is a step, not a reminder. Every citation, figure and quotation produced by a tool is checked against the source before it leaves the firm, and the check is recorded.
- The source travels with the output. Tools that link each claim to the document it came from make verification a click rather than a search.
- Supervision is named. A qualified person owns the output, knows AI was used, and is the one who signs.
- Disclosure is the default. Internally, always. To the court or client, where required, without waiting to be asked.
5. The wider lesson
The sanctions ledger is often read as a story about lawyers and chatbots. It is better read as the first large, public dataset on what happens when professionals adopt tools faster than they adopt verification. Every profession is about to generate its own version. The firms that will not appear in it are the ones treating verification as workflow design rather than as a memo.
Hive Newsroom follows what is changing in AI and professional services. Sources are linked in the text; figures are as published at the time of writing. Comments and corrections: press@get-hive.ai. Back to the .