Claude AI introduces a self-hosted sandbox in public beta for Claude Managed Agents. Tool execution runs in a user-controlled environment connected to private MPC servers, while the agent loop for orchestration, context management, and error recovery remains on Anthropic infrastructure. Network policies, audit logging, and security tooling apply, and files and repositories stay within the user perimeter. Users control compute sizing and the runtime image for compute-heavy work. Separately, a security guidance plugin for Claude Code scans for vulnerabilities during file edits, after AI-generated changes, and at commit time. It analyzes risky code patterns, reviews full diffs, and checks surrounding context, aiming to catch issues before full code review. Internal rollout and benchmarks show a 30–40% decrease in security-related PR comments.
"Claude Managed Agents can now operate in a user-controlled sandbox connected to the user's private MPC servers. Tool execution moves to an environment you configure-your own infrastructure or a managed provider like Cloudflare, Daytona, Modal, or Vercel-while the agent loop that handles orchestration, context management, and error recovery stays on Anthropic's infrastructure. Your network policies, audit logging, and security tooling apply, files and repositories don't leave your perimeter, and you control compute sizing and the runtime image for compute-heavy work."
"The plugin scans for vulnerabilities on file edits, after AI-generated changes, and at commit time, analyzing risky code patterns, reviewing full diffs, and examining surrounding context. Available through the official Anthropic marketplace, the plugin has been widely used internally by the AI company. Across our internal rollout and benchmarks, we've seen a 30-40% decrease in security-related comments on PRs opened using the plugin. The plugin serves as a lightweight first pass, catching issues before a full code review."
Read at SecurityWeek
Unable to calculate read time
Collection
[
|
...
]