Court Rebuke Highlights Risks of Relying on AI like Claude for Legal Briefs
A recent court opinion criticized a lawyer for submitting a brief riddled with errors generated by Claude AI.
Why it matters: Legal professionals are increasingly integrating AI tools into their work. But high-profile judicial criticism underscores the risk of AI hallucinations, ethical pitfalls, and a need for vigilant oversight.
- On May 19, 2026, a court opinion criticized a lawyer for blindly trusting AI-generated content from Claude despite factual errors.
- In February 2026, Judge Jed S. Rakoff ruled that documents created with Claude lacked attorney-client privilege in United States v. Heppner.
- Courts have found that AI platforms do not establish attorney-client relationships, nor ensure confidentiality for users.
- In May 2025, a magistrate judge struck expert testimony after Claude produced a citation to a non-existent article.
A recent opinion reported by Volokh Conspiracy spotlighted a lawyer who "blindly and solely trusted Claude" to generate content for a legal brief. The document contained factual errors, prompting judicial criticism of the attorney's unchecked reliance on artificial intelligence.
- "How [plaintiff's lawyer] then could have blindly and solely trusted Claude to remedy the brief is difficult to fathom," the court observed.
Legal professionals' integration of AI like Claude is creating new challenges. In United States v. Heppner, Judge Jed S. Rakoff determined that 31 documents generated by the defendant using Claude were not protected by attorney-client privilege or the work product doctrine. The court noted:
- "In the absence of an attorney-client relationship, the discussion of legal issues between two non-attorneys is not protected by attorney-client privilege."
- Claude's privacy policy allowed for data retention and sharing with third parties—further undermining confidentiality expectations.
AI "hallucination" remains a documented problem. In May 2025, Magistrate Judge Susan van Keulen struck part of an expert's declaration in Concord Music Group, Inc. v. Anthropic PBC because Claude cited a non-existent article:
- "It is not clear how such an error... could have escaped correction during manual cite-check by a human being."
Recent analysis by Stanford University found that large language models hallucinate legal facts 69% to 88% of the time, amplifying the risk if lawyers forgo due diligence.
By the numbers:
- 31 — Number of documents generated with Claude in Heppner deemed not privileged
- 69%–88% — Rate at which large language models hallucinate legal facts, per Stanford study