Anthropic Open-Sources Sandbox-Runtime, Restricting AI Agents' Access to SSH Keys and Credentials
Sign in✓ Signed in
Anthropic has released sandbox-runtime (srt), an open-source OS-level sandboxing tool designed to prevent AI agents and MCP servers from accessing sensitive credentials by default. The toolkit, released October 24, 2026, under Apache-2.0 license and available on GitHub, addresses a fundamental supply-chain security risk: every MCP server running locally—from GitHub to Slack to third-party tools—inherits full filesystem and network permissions, allowing it to read SSH keys, .env files, and credential stores without restriction.
Continue reading
The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in✓ Signed in — this article isn’t included in your current plan.
Unlocking the full article...