Claude Cowork Security Risks: The Enterprise Guide to Safe Deployment

This guide examines six security attack surfaces for Claude Cowork, Anthropic’s desktop AI agent, including prompt injection vulnerabilities and an audit logging gap where Cowork activity is excluded from compliance tooling. TrueFoundry’s AI Gateway and MCP Gateway are positioned to provide centralized observability, model governance, and tool access controls that native infrastructure does not offer.