Chainguard Harness Uses Claude Mythos 5.1 to Find and Fix Exploitable Bugs

How Harness works

According to Chainguard, Harness connects to a team's repository through the Guardener GitHub app, builds the application exactly as it ships, and boots it in an isolated microVM. A frontier model then probes the running application for exploitable vulnerabilities, testing it in context, the way an attacker would, rather than reading static code.

Every finding is replayed in a fresh, isolated environment until it reproduces every time, so teams only see what is proven. Once a vulnerability is confirmed, Chainguard delivers reproduction steps together with a fix written for that specific codebase. The company says this covers both first-party code and open source dependencies.

Chainguard describes the aim as giving teams frontier-model vulnerability hunting without needing their own access to frontier models, and letting them act only on proven findings rather than sifting through noise.

The problem Chainguard is targeting

Chainguard argues that finding vulnerabilities is no longer the hard part. In its words, frontier AI models have become remarkably effective at finding and chaining exploitable vulnerabilities, but most organizations cannot put that capability to work on their own code. Access to models that perform vulnerability discovery across first-party and open source code has, according to the company, been mostly available to a select group of organizations doing defensive security work for critical infrastructure.

The company also criticizes traditional scanners. They analyze static code, flag issues that may never be reachable, and miss logic flaws, injection paths and resource exhaustion bugs that only appear when an application is running. The result, Chainguard says, is a growing backlog of unverified alerts, with engineers spending hours and tokens triaging findings and writing their own patches instead of shipping.

Dan Lorenc, CEO and Co-founder of Chainguard, said: "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."

Safeguards and launch model

Chainguard lists several design choices around safety and quality:

• Isolation: every run happens in its own microVM sandbox built from source on Chainguard OS, with no network egress, no credentials inside, and full audit logging.
• Human review: a Chainguard engineer reviews and approves every fix and report before it reaches a customer.
• Responsible disclosure: vulnerabilities found in open source go upstream to maintainers through Athena, Chainguard's coordinated vulnerability disclosure process.

At launch, Chainguard engineers operate Harness on behalf of customers. They run it against the trusted open source artifacts Chainguard already distributes and deliver patched builds via Chainguard Repository. They also run it against a customer's source code and deliver an engineer-reviewed report of validated findings, including severity, impact, reproduction steps and recommended remediation.

Purchase through the Claude Marketplace

Harness is available in the Claude Marketplace, which lets enterprises use a portion of their existing Anthropic commitment to buy it. Anthropic manages invoicing, so Anthropic customers can add Harness without a separate procurement process and consolidate their AI spend.

Daniela Hurtado, Head of Commercial Cyber at Anthropic, said that models like Claude Mythos 5.1 "can surface vulnerabilities that have gone unnoticed for years, so how that capability is deployed matters as much as the capability itself." She added that Harness "surrounds Mythos 5.1 with isolation, reproduction, and human review."

What it means for security teams

For organizations without direct access to frontier models, Harness offers a managed route to AI-driven vulnerability discovery, with output limited to reproduced findings and reviewed fixes. Because Chainguard operates the tool at launch, teams receive reports rather than running the model themselves. Availability is currently limited to a beta waitlist, and the source gives no pricing. More information is on Chainguard's Harness page at www.chainguard.dev/harness.