AI filings are getting lawyers disciplined for a reason
The lesson from this week’s sanction cluster is simple: if AI drafts it, somebody must verify it before it leaves the building.
An AI-assisted editorial, reviewed by a human before publishing. It reasons over our own tracker data (and cited context) — a point of view, not legal advice.
A junior solicitor filed false citations generated by an AI tool in Google Scholar, and nobody checked the work. That is the whole story in Murphy, and it is enough to get a lawyer in trouble without adding any dramatic moral about the machines.
For a practicing lawyer, the point is not that one person made one sloppy filing. It is that courts are now seeing the same failure mode over and over: AI output, no verification, discipline. That is no longer a curiosity tucked into a training deck. It is a procurement problem, a supervision problem, and a file-management problem, which is to say a law-firm problem.
If your process cannot prove verification, your process is the defect.
The pattern is already visible
This week’s cluster makes the pattern harder to ignore. In Uprise, Jan Tomasik and Daniel Mann filed a brief with bogus citations generated by ChatGPT, and Judge Hardy ordered them removed, referred them to the Nevada bar, and fined them. In Park v. Kim, Walters filed papers with hallucinated citations and the court sanctioned him for the fake authorities. In California, one attorney was fined $10,000 after filing ChatGPT-generated fake case quotes and citations.
Those are different lawyers, different courts, different levels of consequence. Same basic breakdown. The tool produced text that looked like law; the lawyer treated it like law; the court treated that as the problem. If there is a lesson a billing partner can understand without a memo, it is that the review step is not optional just because the first draft arrived quickly.
The old excuse is gone
The more interesting part is that the law is no longer waiting for perfect proof that a lawyer knowingly tried to fool the court. The verified record already includes Steven Schwartz and Peter LoDuca submitting a ChatGPT-produced brief with fictitious cases, a California appellate sanction for fake quotes and citations, and a 2023 federal matter in which two lawyers and their firm were fined $5,000 each after one used ChatGPT. It also includes Harber, where a litigant in person filed nine non-genuine authorities hallucinated by generative AI, and a UK tax tribunal case where an appellant filed nine bogus tribunal decisions and said she may have used ChatGPT.
That is the useful warning for in-house counsel. Courts are not asking whether AI is morally suspicious. They are asking whether the filing was checked. Once the answer is no, the source of the text stops mattering. An attorney who assumes the software will behave like a junior associate with a conscience is already doing risk analysis by hope.
The market is catching up, slowly
Outside the courtroom, the wider market is moving in the same direction. UC Davis’s law library says courts, judges, and state bar associations are actively developing ethics rules, opinions, local rules, and guidelines on AI use in legal practice. Bloomberg Law reported that recent IP decisions send a consistent message: federal courts will not prohibit AI use, but they will sanction lawyers who use AI tools without meeting their ethical obligations, and liability can reach all signing attorneys. Mondaq reported that the California State Bar published new guidance in 2026 saying AI may assist lawyers, but it may not displace duties of competence, diligence, confidentiality, candor, supervision, communication, and compliance with law.
That is the part procurement teams need to hear. A policy that says lawyers should “use judgment” is not a control. A real control is line-citing, source checking, assigned responsibility, and a stop sign before filing. Turbo Law, which publishes this site, says its platform proposes and the team verifies, which is at least the correct order of operations. The interesting part is not the software label. It is the admission that the human review must be built into the workflow, because after the filing there is no graceful recovery.
What a sensible firm does now
Pinsent Masons said it took steps to put safeguards on its internal AI pilot after a court incident involving AI-generated errors, and it also piloted Lexis+ AI for legal research, drafting, and review while measuring time savings and user adoption. That sounds unglamorous, which is why it is probably the right model. The firms that survive this phase will not be the ones with the most enthusiastic memos about innovation. They will be the ones that can show who checked what, against which source, before the signature went on.
This is the reckoning for working lawyers right now. AI can speed up first drafts and research, but the bill for a missed citation lands on the signer, not on the prompt. If your process cannot prove verification, your process is the defect.