Roughly 700 documented instances of AI hallucinations in U.S. court filings since early 2025. That number, tracked by an ongoing database, captures just how fast legal AI adoption has outrun the courts' ability to manage it.
A new analysis from the R Street Institute, published March 13, lays out the growing collision between lawyers embracing AI tools and a judicial system that has no coherent framework for dealing with them. The picture is messy: AI is saving thousands of hours in public defender offices while simultaneously generating fabricated case law that gets lawyers fined.
The Privilege Ruling That Should Worry Every Law Firm
In February 2026, U.S. District Judge Jed Rakoff issued what may be the most consequential AI ruling for legal practice so far. In U.S. v. Heppner, a $300 million securities fraud case, the defendant had used chatbots to brainstorm defense strategies and stress-test legal theories. Judge Rakoff ruled those conversations are not protected by attorney-client privilege. His reasoning was blunt: chatbots are not lawyers, so the privilege does not apply.
That one ruling changes the risk calculation for every attorney who has typed a case detail into ChatGPT or Claude. Anything you share with a chatbot could be discoverable in litigation. Judge Rakoff acknowledged AI as "a new frontier" but emphasized that "AI's novelty does not mean that its use is not subject to longstanding legal principles."
Where AI Is Actually Working
The story is not all cautionary. The LA County Public Defender's Office uses AI to read incoming police reports, standardize formats, and extract relevant information, saving thousands of staff hours per year. Montgomery County, Texas prosecutors use it to summarize handwritten documents, translate Spanish to English, and pull data from social media. Over 80% of criminal cases now involve body camera video, and AI tools search that footage for critical moments like Miranda warnings and signs of coercive interrogation.
These are genuine productivity gains in offices that are chronically understaffed. The R Street analysis frames this through the lens of "cost disease," the economic pattern where labor-intensive industries get more expensive relative to the broader economy. AI could be the cure, if the profession can figure out guardrails.
The Guardrails Problem
So far, guardrails have been reactive. A legal team in Alabama was fined $5,000 in late 2025 for submitting AI-fabricated case law. Months later, a Wisconsin team got hit with similar penalties. In Nevada County, California, a defense attorney argued that AI-generated errors represent "an existential threat to due process," which prompted the local DA to appoint an AI policy coordinator and implement office-wide guidelines.
King County, Washington took perhaps the strongest stance: in 2024, Prosecuting Attorney Dan Clark declined to accept AI-assisted police narratives as evidence, citing concerns about blurring the line between software output and officer testimony.
The patchwork approach is the real problem. Every jurisdiction is making up its own rules. Some courts require disclosure of AI use, others do not. Specialized legal platforms like Westlaw AI and Lexis+ AI are more controlled environments than general-purpose chatbots, but even those are not immune to inaccuracies.
For anyone using AI tools in professional settings with legal exposure, the Heppner ruling is the headline to remember. Your chatbot conversations may not be private. And until courts establish consistent standards, the safest approach is to treat every AI interaction as potentially discoverable.