291/69 Thursday, May 28, 2026
Anthropic has introduced new security features for Claude, including a Self-hosted Sandbox for Claude Managed Agents and a Security Guidance Plugin for Claude Code. The Self-hosted Sandbox is currently available in Public Beta and is designed to allow Agent tools or code execution to run within an environment controlled by enterprise users, such as the organization’s own infrastructure or a supported hosting provider. Meanwhile, the Agent orchestration process remains on Anthropic’s systems.
These features aim to reduce the risks associated with allowing AI Agents to directly access an organization’s files, systems, or networks. They enable organizations to apply their own network policies, logging systems, and security tools, while files and repositories remain within the boundaries of an environment controlled by the organization. Anthropic stated that Claude Code’s Sandbox applies both file and network restrictions to help reduce risks from prompt injection, where hidden instructions may attempt to cause an Agent to access files or connect to unauthorized destinations.
At the same time, Anthropic also launched the Security Guidance Plugin for Claude Code to help detect insecure coding patterns while Claude is writing or modifying files. Examples include command injection, cross-site scripting (XSS), and unsafe deserialization. The plugin helps Claude review and fix identified issues within the same session before the code enters the Pull Request stage. However, developers and organizations should still review and perform security testing on the code before deploying it to production.
Source https://www.securityweek.com/anthropic-releases-new-claude-sandbox-security-guidance-plugin/