The AI Slop Machine Is Now Institutional
Sanctions and judicial scolding have failed to stop hallucinated filings because firms treat verification as a nuisance rather than a duty.
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.
Judges are busy, and no practitioner wants to spend an evening re-checking every reporter volume cited in a routine opposition brief. When generative software can spit out thirty pages of apparently plausible precedent in thirty seconds, the temptation to skim the output and trust the software is obvious. But giving into that temptation has metastasized from a handful of embarrassing blunders into an institutional malpractice crisis.
The proof is now documented in a public database. French lawyer and data scientist Damien Charlotin recently launched a repository tracking fake legal citations and arguments generated by AI in court filings, documenting 120 instances since June 2023. Mashable reports that the database logged 36 cases in 2024, rising to 48 in the first half of 2025 alone. Just recently, lawyers representing Anthropic were caught relying on a false attribution generated by Claude in expert testimony. These are not isolated accidents by pro se litigants in obscure municipal courts. They are systemic failures occurring at every tier of the bar.
Courts relied on procedural scolding and local administrative rules, expecting the bar to self-correct once the risk of public humiliation became clear. Instead, the problem escalated.
Three Years of Failed Scolding
We have been down this path before. In 2023, two attorneys in the U.S. District Court for the Southern District of New York submitted a brief containing fabricated case citations generated by ChatGPT, earning a $5,000 fine for each lawyer and their firm. That same year, a South African judge in Parker v Forsyth NO and Others took a gentler approach, issuing a warning after finding no intent to mislead when an attorney presented fake case law produced by ChatGPT.
The bench assumed that early monetary fines and public reprimands would serve as an adequate deterrent. That calculation was wrong. Courts relied on procedural scolding and local administrative rules, expecting the bar to self-correct once the risk of public humiliation became clear. Instead, the problem escalated.
A Florida appellate court panel recently scolded an attorney over filings filled with incomprehensible AI slop that clogs dockets and delays proceedings, warning that the lawyer could face state bar discipline. Across the country, our database shows courts have logged 858 AI-related sanction incidents to date. New York accounts for 67 of those incidents and California leads with 75. The sheer volume of orders demonstrates that slap-on-the-wrist fines have done nothing to stem the tide.
The Illusion of Automation
The structural cause of this failure is straightforward. Law firms treat generative AI as a labor-saving shortcut for grunt work rather than an untrustworthy drafting assistant that requires rigorous line-by-line verification.
Real diligence requires tools and workflows designed specifically to force human oversight. Specialized litigation software like Turbo Law, which publishes this site, hyper-links assertions directly back to source documents because raw language models cannot be trusted to self-correct. When partners treat verification as a clerical afterthought, hallucinated citations inevitably reach the docket.
If three years of sanctions, published reprimands, and mandatory AI disclosures have not stopped lawyers from filing fabricated cases, procedural admonitions never will. Until firms face strict internal liability for filing unverified software outputs, fake citations will remain a routine feature of modern litigation.