Overview
- On February 10, U.S. District Judge Jed S. Rakoff ordered about 31 documents created with Anthropic’s Claude to be produced to prosecutors, ruling from the bench with no written opinion yet issued.
- The court found no attorney‑client privilege because inputs and outputs were shared with a third‑party tool whose terms allowed disclosure and training, defeating any reasonable expectation of confidentiality.
- Work‑product protection was rejected because the materials were generated independently by the defendant and not at the behest or direction of counsel.
- The decision was fact‑specific to a publicly available consumer AI platform and did not resolve how enterprise tools with contractual confidentiality or attorney‑directed AI use might be treated.
- Legal commentators advise updating AI governance to preserve protections by using secured enterprise deployments, limiting dissemination, and routing AI‑assisted legal work through counsel, as the criminal trial remains set for April 6.