Industry newsAi hallucinations
The Hallucination Blame Game Says More About Buyer Diligence Than About Any Legal AI Tool
eDiscovery Today's Doug Austin reported on June 4, 2026 that attorneys are increasingly naming specific legal AI vendors when hallucinated citations turn up in court filings, drawing on a Project Counsel Media/Luminative Media investigation and on Damien Charlotin's hallucination-case database. Charlotin, an attorney and data scientist, told Law360 that only about 10% of entries in his database actually name the tool involved, with ChatGPT the most frequently cited — likely because pro se litigants use it, not lawyers. The scale of the market makes that vagueness worse: LegalTech Hub's March map counted 1,014 GenAI product placements from 806 vendors, more than double the 400 placements it tracked in February 2025, while other reporting cites attorneys collectively now using over 800 different legal AI tools.
How often is a hallucination actually traced to a named vendor?
Rarely — Charlotin's own database, the most cited source for these incidents, attributes a specific tool in only about one in ten entries.
That gap matters for anyone reading a headline about a vendor named in a court order and assuming the industry has a measurable performance problem. It doesn't, not reliably. Charlotin himself noted that lawyers running a matter through several tools can pick which one to name once an error surfaces, which means the public tally of vendor-attributed hallucinations reflects disclosure choices by counsel as much as it reflects the tools themselves.
Why did vendors stop promising zero hallucinations?
Most major legal AI providers have quietly dropped hallucination-free guarantees, leaning instead on retrieval-augmented generation and citation-checking to reduce, not eliminate, the error rate.
That's worth noticing precisely because RAG has become the industry's default pitch, and RAG is only as defensible as the retrieval layer behind it — what corpus it's grounded in, whether every generated answer actually links back to a real document, and whether that link gets checked before it reaches a filing. A tool that retrieves accurately but still lets a lawyer paste an unverified answer into a brief hasn't solved the problem; it's just moved the failure point downstream.
Is the market really flooded with 800-plus tools, and does that change the risk?
Yes — LegalTech Hub counted 1,014 GenAI product placements from 806 vendors in March, more than double its February 2025 count of 400.
A market that grew that fast in roughly a year is one where due diligence hasn't caught up with procurement. Firms trialing research or drafting tools without asking how each vendor sources case law, or what happens when a citation can't be verified, are setting up the exact scenario Charlotin's database is now cataloguing.
Who actually bears responsibility when a citation is wrong?
Courts and commentators are converging on the same answer: the filing attorney, not the software, because verification was always counsel's job.
Northwestern's Daniel W. Linna Jr. put it plainly, saying that being named in a court order damages a vendor's reputation but that "savvy lawyers know that the errors say a lot more about the lawyer than the platform." That lines up with sanctions orders going back to the earliest ChatGPT filing incidents: professional responsibility rules have never carried a tool-liability exception.
Frequently asked questions
Does using a RAG-based tool guarantee citations are real?
No. RAG lowers the odds of fabricated case law by grounding answers in retrieved documents, but a lawyer still has to confirm each cited source exists and says what the tool claims before filing it.
Should a firm avoid naming its AI vendor if a hallucination happens?
That's a candor question for the court, not a strategy question. Charlotin's data suggests many lawyers already omit the tool name, which only obscures which products need scrutiny.
What should a buyer ask a vendor before adopting a research or drafting tool?
Ask how each answer is grounded and cited, what happens when a source can't be verified, and whether the vendor still avoids promising zero hallucinations.
Sources: eDiscovery Today, reporting by Doug Austin, drawing on findings from Project Counsel Media/Luminative Media and Law360.
The original report
Industry headlines from other publications. Each links to the original reporting on the publisher's own site.