Lawyer AI Sanctions Surge as Firm Adoption Outpaces Training

Fake citations have become a routine courtroom problem

When two New York lawyers were fined $5,000 in June 2023 for filing a brief built on cases ChatGPT had invented, the episode looked like a one-time embarrassment. It now looks like the start of a trend. Reuters-sourced reporting this autumn says dozens of lawyers have been sanctioned for filing briefs in which AI made up case citations or misquoted the law.20 The cases are also getting worse. In September, the New Mexico Supreme Court fined defense attorney Stephen Aarons and held him in contempt over a murder appeal brief that contained police and witness testimony ChatGPT had fabricated.20 Aarons said he had given a computer-generated transcript to the chatbot, expecting what he called a "bulletproof summary," and had not understood how badly AI can hallucinate facts.20

The New Mexico case stands out because it goes beyond the usual failure. Most sanctions so far have involved invented case law. Here the invented material was factual testimony in a criminal appeal where the client was serving a life sentence.20 That difference matters for the industry, because it shows the risk is not limited to legal research. It reaches any task where lawyers ask a general-purpose model to summarize the record.

How big is the problem? It depends who is counting

Nobody agrees on the size of the problem, and the gap between estimates says something about how quickly it is growing. One analysis of the public AI Hallucination Cases database, maintained by researcher Damien Charlotin, puts the worldwide total above 2,100 court decisions involving authority an AI invented or misstated. Where an outcome was recorded, 379 led to a monetary sanction and 162 to a disciplinary referral.11 Another tracker put the figure at roughly 1,490 worldwide, with more than 1,000 in the US, as of May 2026, and noted that most involve self-represented litigants rather than lawyers.12 A separate count published in October cites 1,395 cases.19 Illinois court officials, citing a Bloomberg Law analysis, reported in January that more than 280 filings had included hallucinated citations since 2023, and that such cases grew sevenfold in 2025.16

The numbers differ because they were taken at different times and count different things, but they all point the same way. Mata v. Avianca was one of six such decisions logged in its quarter, and courts now log several hundred a quarter.11 The headline-grabbing fines are also less typical than the coverage suggests. Of the 191 US-dollar penalties in the database, the largest group, 86, falls between $1,000 and $5,000. Only six exceed $50,000.11 That database also covers penalties against all filers, not just lawyers.11

Penalties are escalating

Even so, the upper end of the range has risen sharply. Law.com reported a record $110,000 fine this year against two lawyers who filed fake citations and quotes.15 In March 2026, the Sixth Circuit ordered each of two attorneys in Whiting v. City of Athens to pay $15,000 in punitive sanctions, plus the other side's attorney fees and double costs, after finding more than two dozen fake or incorrect citations.16 In April, the Nebraska Supreme Court indefinitely suspended an Omaha lawyer whose divorce appeal brief had 57 defective citations out of 63. He initially denied using AI.1215 In June, the Ninth Circuit suspended two immigration lawyers from practicing before it for six months and ordered every attorney at their firm to certify AI use and verification on filings for two years.19

The problem is not confined to the US. In September 2026, England's Solicitors Disciplinary Tribunal struck off a registered foreign lawyer over AI-generated fake cases, the first AI misuse case litigated before that body.16 Regulators in Australia have restricted a solicitor's practising certificate, and a court in the Philippines has warned lawyers against overreliance on AI.16

Courts are also moving from punishment to procedure. Four Cherokee County, Georgia, judges have issued a standing order requiring anyone who uses AI to prepare a filing to disclose it, certify that the citations are accurate, and name the tool.18 The order followed a Clayton County murder appeal in which a prosecutor admitted that AI produced at least nine fake or misattributed citations. Some of those errors ended up in the trial court's own order.18 Prosecutors are not exempt, and a judge's signature does not catch everything.

Adoption has outrun governance

For the legal tech industry, the key context is how fast the profession has taken up these tools. 8am's 2026 Legal Industry Report, based on more than 1,300 legal professionals, found that 69% now use general-purpose generative AI for work, up from 31% a year earlier.8 ILTA's 2026 Technology Survey of 508 firms found that 94% are using or exploring generative AI, up from 80%.2 Thomson Reuters reports that 41% of law firms say their legal teams use GenAI, up from 28%, and 74% of users apply it to legal research.3 Everlaw's litigation-focused survey puts active use at 49%.7

The adoption figures vary even more than the sanctions counts, mostly because survey questions differ. One review notes that, depending on what counts as "using AI," estimates run from 26% to 92%.9 The trend, however, is clear.

The weak point is governance. In the 8am data, 54% of respondents said their firms offer no AI training and have no plans to start, and only 11% reported mandatory training.8 Forty-three percent said their firm has no formal AI policy and no plans to create one, and only 9% described a written policy that is actually enforced.1 Everlaw found that only 34% of respondents are integrating GenAI across their organization, while the rest rely on piecemeal approaches by partner or practice group.10 Only 19% of legal professionals call their firm "very prepared" for AI-driven change.8

The two sets of data fit together. Individual lawyers have adopted consumer chatbots faster than their firms have set rules for them, and the sanctions record is what that gap produces. Aarons's explanation, that he did not understand AI could invent facts, is the kind of mistake basic training is meant to prevent.20 An analysis of lawyers' own explanations found the same pattern repeatedly: deadline pressure, a new tool treated like a search engine, and an assumption that a properly formatted citation must have come from somewhere.11

Legal-specific tools do not remove the risk

The industry's main pitch is that purpose-built legal AI fixes the problem consumer chatbots create. There is some support for that. In the 8am survey, 43% of firms using legal-specific tools said they trust the output more than consumer options.8 But the record does not show that buying a legal tool solves the problem. In Wadsworth v. Walmart, Morgan & Morgan attorneys cited nine cases, eight of them fake, after a lawyer used the firm's in-house AI database.12 Another reported case involved a brief drafted with Westlaw's AI research platform and CoCounsel that still contained hallucinated citations.17 Independent benchmark testing found that 24% of 3,000 graded frontier-model legal answers cited or applied law that does not support the claim.9

Much of the detailed analysis of these cases comes from companies selling legal AI, and they have an obvious interest in describing the problem as one of "ungrounded" generic tools.1117 Their data is useful, but it should be read with that in mind. The court decisions themselves point elsewhere. The Ninth Circuit put it plainly: the rules are broken not during research and drafting, but when a lawyer signs and files.19 No vendor can take over that duty. Bednar in Utah, Crabill in Colorado, and the lawyers in Mata all tried to shift blame to a clerk, an intern, or the software, and the database shows those defenses rarely work.1112

Larger fines alone won't fix this

The most likely conclusion is that the profession is depending on deterrence to fix a problem that requires institutional controls. Fines have risen from $5,000 to six figures, yet the caseload keeps growing.1511 The ABA Journal summarized the Law.com reporting by saying fines alone are not enough.15 One commentator made a similar point: sanctions act on a lawyer after the damage is done, not on the work before it is filed, and a lawyer facing a deadline may reasonably bet on not getting caught.19

The changes most likely to matter are the less visible ones, such as disclosure-and-certification orders like Cherokee County's, firm-wide verification requirements like the Ninth Circuit's, and training that firms are not yet providing.18198 Cost and return on investment have now entered ILTA's top three generative AI concerns at 41%, while worries about accuracy have dropped.2 Given the sanctions record, that shift looks premature. Firms that budget for AI licenses but not for verification are taking the same risk that has led to the sanctions described above.