Chainguard Adds AI Vulnerability Hunting and Coding Agent Tools

Chainguard introduced a service for testing running applications with frontier AI models, along with tools for isolating coding agents and checking their output, as it expands security coverage for Go and .NET packages.

The company announced the offerings Oct. 14 at its Assemble London conference. Harness uses models such as Anthropic’s Claude Mythos 5.1 to probe applications built from source in isolated microvirtual machines.

Harness reproduces reported exploits in a fresh environment before generating a finding. It also checks recommended fixes against the exploit. The service is accepting beta waitlist sign-ups, and organizations can purchase it through Anthropic’s Claude Marketplace using part of an existing Anthropic commitment.

Chainguard Workspaces provides remote, hardened microVMs for developers and coding agents. Developers can sync local working trees, including uncommitted changes, to a workspace, while credentials stay outside the virtual machine.

Chainguard Checks runs continuous integration tests on code produced in those workspaces. A workspace can pause while checks run, then resume with failure logs available to the agent. Both Workspaces and Checks are in beta.

Chainguard also expanded its Libraries service to scan Go and .NET packages for malware and greyware. The new coverage is in early access. During beta, the scanner returned malware verdicts on more than 5,000 packages across the two ecosystems, with more than 10% of those verdicts lacking a match in public sources at detection time.