AI company Anthropic announced two new security features at this week’s "Code w/ Claude" event in London: a self-hosted sandbox and a security guidance plugin. The sandbox is currently in public beta, while the plugin is already widely used internally at the company. These new tools aim to help developers code with Claude AI in a more secure environment while also improving code safety.
According to Anthropic, Claude Managed Agents can now run in a user-controlled sandbox that connects to the user’s private MPC server. This design allows enterprises to confine AI agent operations within a controlled environment, reducing the risk of data leaks and unauthorized access. Meanwhile, the security guidance plugin for Claude Code can detect and fix vulnerabilities in real time as developers write code. The plugin is available for download through Anthropic’s official marketplace and has been thoroughly validated in the company’s internal projects.
