Most firms have, at minimum, experimented with AI tools for legal research, drafting, or summarization. That is not surprising. What is surprising is how quickly this has become a documented source of court sanctions, and how directly it connects to the kind of exposure professional liability coverage is meant to address.
The Scale of the Problem
A database tracking AI-related errors in court filings has now logged more than 1,400 cases over the past three years where courts addressed AI-generated content, including fabricated case citations, invented quotes, and arguments built on cases that do not exist.
The pace has accelerated sharply. Over 300 of these incidents were documented in the period since mid-2023, with roughly 200 occurring in 2025 alone. This is not confined to one jurisdiction or one type of practice. Cases have surfaced in federal and state courts across the country, and in courts in the UK, Australia, Canada, and Israel.
In the first two weeks of August 2025 alone, three separate federal courts issued sanctions against attorneys for AI-generated hallucinations in filings. One of those cases involved an attorney using a well-known commercial legal research platform that itself produced fabricated citations.
How These Cases Actually Play Out
The pattern across documented cases tends to follow a similar arc. An attorney uses an AI tool to help with research or drafting. The tool produces citations that look entirely plausible, complete with case names, docket numbers, and even the names of real judges, but the cases themselves do not exist or do not say what the filing claims they say.
In one widely reported matter, a federal judge found that 12 of 19 cases cited in a brief were fabricated, misleading, or unsupported. The filing was described as "replete with citation-related deficiencies, including those consistent with artificial intelligence generated hallucinations."
What stands out in several of these cases is what happens after the first error is caught. In one Alabama Supreme Court matter, an attorney was sanctioned for citing a made-up case, told the court it would not happen again, and then cited another nonexistent case in the very next sentence of the same filing.
Courts have started distinguishing between attorneys who disclose AI use candidly and correct the record quickly, versus those who deny using AI when the evidence suggests otherwise. The remedial steps that tend to reduce the severity of sanctions include withdrawing the problematic filing immediately, being transparent about what happened, and in some cases covering opposing counsel's wasted time. Attorneys who deny AI use or fail to correct errors after being warned face significantly harsher outcomes, including referrals for bar discipline.
Why This Matters for Professional Liability Coverage
A sanctions order is a matter of public record. It is the kind of event that can prompt a client to ask questions, and in some cases it directly affects the underlying matter. If a brief gets struck, an argument gets discredited, or a case gets delayed because of fabricated citations, the client has a real argument that their case was harmed.
This is a relatively new fact pattern. Five years ago, this type of error simply did not exist as a category. Professional liability policies were not written with this specific risk in mind, and the legal industry as a whole, including courts, bar associations, and insurers, is still working out exactly how it gets treated.
What is becoming clearer is the standard of care question. Courts have made clear that even unintentional reliance on AI does not excuse an attorney from responsibility for what they file. The signature on the brief is the attorney's, regardless of what tool helped produce it.
What This Means Practically
This is not a reason to avoid AI tools. Adoption is widespread and likely to keep growing. It is a reason to treat AI output the same way you would treat a draft from a first-year associate or an outside contractor: useful, but not something that goes into a filing without independent verification.
A few practical points worth considering for any firm using AI tools in client work:
Every citation in a filing should be independently verified against the actual case, not just checked for the existence of a case with that name. Fabricated citations are often built to look exactly like the citation a researcher would expect to find, which is part of why they slip through.
Some courts now require certifications that AI-generated content has been checked for accuracy by a human being. Even where not required, having an internal policy that documents this verification step protects the firm if a question ever arises.
If an error does make it into a filing, the documented pattern is clear: catching it quickly, disclosing it candidly, and correcting the record produces a far better outcome than denial or delay.
This is a fast-moving area, both in terms of how courts are responding and how the insurance market is thinking about it. If your firm has questions about how AI use fits into your current coverage, it is a conversation worth having before it becomes relevant rather than after.
Sources: Jones Walker LLP, "From Enhancement to Dependency: What the Epidemic of AI Failures in Law Means for Professionals" (August 2025); Scientific American, "Why Lawyers Keep Citing Fake Cases Invented by AI"; database of AI hallucination cases maintained by Damien Charlotin, HEC Paris.