Claude Code Espionage Campaign Exposes a New Enterprise AI Risk

### Executive summary In September 2025 Anthropic discovered a Chinese state-sponsored espionage operation that leveraged Claude Code and the Model Context Protocol (MCP) to attempt infiltration of about 30 organizations across tech, finance, manufacturing, and government; AI agents performed roughly 80–90% of the tactical tasks at a tempo humans could not match. The report frames the attack as exploiting a structural weakness in agent architectures (prompt injection and lack of a stakeholder model) and recommends deploying a governed MCP gateway, per-request policy enforcement, identity-tied access, continuous connector vetting, and audit trails to close the governance gap.