Anthropic Compliance API Improves Claude Code Security Visibility

Anthropic has added local session transcript endpoints to its Compliance API, giving security teams a clearer audit trail for Claude Code activity. Unlike a conventional chatbot, Claude Code can read local files, execute shell commands, invoke MCP servers, and use the credentials available on a developer’s machine. This architecture separates the model’s cloud-hosted reasoning from the local harness that performs actions.

Token Security reports that local agents account for 68.6% of the AI agents it discovers in customer environments. That deployment model creates visibility and identity challenges because Claude Code has no centralized enterprise console covering local configurations, permissions, and runtime behavior. Historically, limited native monitoring forced organizations to rely on third-party extensions to track activity performed through local harnesses.

The new transcript endpoints can help security teams record what an agent did, but activity logs cannot establish whether its access was legitimate. Teams still need to map each developer, endpoint, credential, MCP integration, plugin, and skill against IAM and least-privilege policies. Because the harness executes commands with the user’s local privileges, enterprises should also audit exposed services with a port scanner, check inherited secrets with a password checker, and verify API connections with an SSL/TLS checker.

A Token Security-commissioned Cloud Security Alliance survey of 418 IT and security professionals found that 68% rated their visibility into AI agents as high. Anthropic’s Compliance API is a significant step toward auditing local-agent behavior, but effective governance will require transcript data to be combined with endpoint telemetry, identity context, credential controls, MCP inventories, and command-level policy enforcement.