Chainguard launches Harness for AI vulnerability testing

Chainguard announced in a press release Harness, a security service that uses Claude Mythos 5.1 from Anthropic to find exploitable vulnerabilities in application code. Harness is available through a beta waitlist and the Claude Marketplace, without requiring customers to have direct access to the model.

Harness connects to a repository through Chainguard's Guardener GitHub app, builds the application in an isolated microVM, and asks the model to probe the running software. The environment has no network egress and no credentials inside, and each finding is replayed in a fresh environment to confirm it can be reproduced.

Chainguard provides reproduction steps and a proposed fix for each confirmed vulnerability. A Chainguard engineer reviews every report and fix before delivery, while vulnerabilities found in open source software are sent to maintainers through the company's Athena disclosure process.

Chainguard launched three related products the same day. Workspaces gives every developer and coding agent its own hardened microVM with no access to sensitive credentials and optional egress policies, so agents can run autonomously while keeping a signed record of what they did. Checks is a continuous verification engine that runs CI as a tool call inside agent development loops and returns signed results in seconds. Chainguard Libraries now covers Go and .NET dependencies, adding CVE remediation for JavaScript and Go and same-day fixes for zero-day vulnerabilities.