A Bloomberg Law report published October 22, 2026, says law-firm leaders raised concerns about contract terms, email access and training after an August 2026 meeting in Nashville. Five attendees spoke anonymously about the private meeting. Anthropic’s published policies describe separate controls for data training, retention and connector permissions, while one firm’s pilot account offers a mixed view of Claude’s use in legal work.
What law-firm leaders raised at the Nashville meeting
Attendees told Bloomberg Law that Anthropic would not negotiate certain privacy and data-security terms without a minimum spending commitment. They also said they felt largely responsible for training lawyers to use Claude. One attendee objected to granting the service email access.
The concerns came as firms considered how to deploy Claude, not as a blanket rejection. Bloomberg Law reported that 44% of firms in an International Legal Technology Association survey were using or exploring Claude. Among that group, 7% said they had fully rolled it out to attorneys.
Anthropic separates training, retention and safety review
Anthropic says customer content from Claude for Work, Claude Platform and the API is not used to train its models by default. That policy addresses model training; Anthropic describes retention and safety review separately in its legal data-policy guidance.
Claude for Work administrators can set an organization-wide retention period, Anthropic says. Zero data retention (ZDR) for the API applies only to eligible endpoints and models. Anthropic also says it retains prompts and outputs for its most capable models for 30 days for safety monitoring.
Anthropic says human reviewers can access content flagged by automated safety systems. It describes additional confidentiality and access controls for that limited review, and says human access is logged for customers using Access Transparency. Anthropic says the appropriate configuration depends on the jurisdiction, client and matter; its guidance is not legal advice.
Email connectors can allow reading while blocking sending
Team and Enterprise owners can restrict connector actions, including allowing email search and summaries while blocking message sending, according to Anthropic’s connector documentation. Connectors inherit a user’s permissions in the underlying service, and administrators can require approval for actions.
Connected services also process data on their own infrastructure under their own terms. Anthropic says those services may process data outside the United States, so connecting Claude to a firm’s email or matter system involves the connected provider’s data practices as well.
Anthropic describes more than 20 connectors and 12 practice-area plugins for legal work. Its announced plugins can be customized with firm playbooks, and its product demonstration presents integrations including Thomson Reuters, iManage, NetDocuments, Docusign, Ironclad and Relativity.
Ogletree Deakins’ pilot combines praise with a reservation
Timothy Fox, chief data and AI officer at Ogletree Deakins, said the firm had been piloting Claude since February 2026. He described the deployment as one of the firm’s most successful and said Claude was useful for large datasets and prototyping. Fox also said Claude was not designed specifically for legal work and that its enterprise security was not designed like a legal platform’s.
Support and training were another concern in the reported accounts. Alex Finkel, chief innovation officer at Offit Kurman, said smaller firms may need ready-made workflows, trainers and AI expertise to put tools like Claude to work. Anthropic’s legal offering includes plugins and connectors, while the meeting attendees described firms taking on much of the training themselves.
