Chainguard Unveils Claude Mythos 5.1 Vulnerability Hunter and AI Agent Sandboxes

Chainguard unveiled four security updates at its Assemble London conference, highlighted by Chainguard Harness, a waitlist beta vulnerability-hunting service that uses frontier AI models like Anthropic's Claude Mythos 5.1 to probe live application builds inside isolated virtual machines. Anthropic is also selling Harness on the Claude Marketplace, allowing enterprise customers to apply existing Anthropic spending commitments toward the staff-operated service, where Chainguard engineers verify every reproducible bug report and fix. To secure AI coding agents earlier in development, the company launched beta tests for Chainguard Workspaces and Chainguard Checks, isolating agents in remote virtual machines without real credentials while running continuous integration tests from warm snapshots before pull requests are opened. Chainguard also expanded its vetted open-source dependency service, Chainguard Libraries, to five languages with early access for Go and .NET alongside Java, Python, and JavaScript. Founded in 2021, Chainguard has raised about $892M in venture funding, including $280M from General Catalyst's Customer Value Fund last October. "Every time we've pointed Harness at code, it has found serious, exploitable vulnerabilities, including in software that's been scanned many times over and code that runs critical infrastructure." — Dan Lorenc, Co-founder and CEO of Chainguard.