Introducing Chainguard Harness: Vulnerability hunting and remediation powered by Claude Mythos 5.1, no model access required

Today, we're introducing Chainguard Harness, a new offering that uses frontier AI models like Anthropic’s Claude Mythos 5.1 to hunt down exploitable vulnerabilities in an organization's application code and create a validated, reproducible report with recommended fixes. Alongside the introduction of Harness, we’re also announcing its availability through the Claude Marketplace by Anthropic, allowing organizations to put a portion of their existing Anthropic commitment toward finding and fixing exploitable vulnerabilities.

We’re opening the beta waitlist for Chainguard Harness today.

Frontier models that perform vulnerability discovery across first-party and open source code have been mostly available to a select group of organizations performing defensive security work for critical infrastructure. Chainguard Harness is now bringing a frontier harness to organizations that lack access or the security and engineering expertise to do so in the face of growing threats from open weight models.

Frontier vulnerability hunting, run by Chainguard

The Chainguard Factory already takes the open source software the world relies on and safely rebuilds it from scratch in hardened, isolated environments. Chainguard Harness applies the same level of rigor to frontier vulnerability discovery and remediation. We designed a custom harness around these advanced AI models. Rather than just scanning static code, Chainguard Harness builds the application from source, and the model actively attacks the live, running instance of your app, exactly like a real attacker would. Here's how the process works:

1. Connect your repositories. Use the Chainguard GitHub App to share repositories with Chainguard, granting read-only access to the repositories, or upload a source archive.

2. Rebuild in isolation. Chainguard Harness builds your application from source exactly as it would ship to production and boots it up inside a secure, isolated microVM.

3. Hunt for exploits. Mythos 5.1 actively probes the running application, looking for vulnerabilities such as logic flaws, injection, and resource exhaustion. It works from your source, git history, dependency source and build configuration, and watches how the live application responds.

4. Prove every finding. Every exploit is replayed in a fresh microVM. If it doesn't land every time, it doesn't make the report. We also confirm that each recommended fix closes the vulnerability: the exploit has to succeed against your current code and fail against the fix.

5. Deliver the fix. Chainguard provides you with reliable steps to reproduce the issue alongside a custom fix written specifically for your codebase. You get a written report covering:

   • An executive summary

   • The severity and impact of each finding

   • The steps to confirm each finding

   • A recommended fix written for your codebase

Isolated by design

Pointing a frontier model at real code demands a harness built for containment, which Harness managed in two distinct phases. The first is a setup phase, which involves a separate model building your application, fetching its dependencies and tooling, and getting it running inside the microVM. The microVM is then suspended. Next comes the hunting phase. The microVM resumes for Mythos 5.1 with no network egress and no credentials inside. The model works only with a shell and packages built by Chainguard.

The agent loop runs outside the sandbox, so every tool call and command the model issues is logged and observable. We also treat the code under test as untrusted data, which guards against prompt injection hidden in the code being analyzed. Your report describes each issue and how to confirm it, not a working attack.

Beyond its approach to containment, Chainguard Harness combines a broad set of engineering innovations to safely and securely discover and help fix exploitable vulnerabilities:

• Live application testing: Evaluates active, running behavior rather than static code to uncover real-world runtime flaws and logic bugs in both first and third-party code.

• Reproducible findings: Every exploit is replayed until it lands every time, and every recommended fix is checked against the exploit it's meant to stop. Your team only spends time on what's real.

• Bespoke fixes across open source and first-party code: Custom recommendations and reporting across the complete code base give security and engineering teams a path to secure code, rather than a list of risks. 

• Human review on every fix: A named Chainguard engineer reviews and approves every fix and report before it reaches you.

• Upstream disclosure: If we discover a flaw in an open source dependency rather than your own code, we handle the coordinated disclosure process directly with the upstream maintainers. Once it's resolved upstream, the patched build ships seamlessly via Chainguard Repository.

Peace of mind without the resource burden

Chainguard Harness gives security and engineering teams the visibility and actionability of frontier AI scanning without the resource, expertise, or risk burdens of managing it in-house:

• Catch risks before attackers do: Proactively find and patch exploitable vulnerabilities across your custom code and open source dependencies, including those runtime bugs that traditional static scanners miss.

• Focus engineering time on actual problems: Reproducible findings focus team attention and resources on the most actionable issues, and since every fix is written specifically for your code, engineers can review and ship the patch instead of researching it from scratch.

• Get top-tier results without the massive costs: You don't need to secure API access to expensive models, build and secure your own testing harness, hire AI specialists, or burn cash on token fees while trying to get things working.

Powered by Claude, available on the Claude Marketplace

Chainguard Harness is powered by Anthropic's most advanced models like Mythos 5.1, and built for safe and responsible use. Only Chainguard staff operate the model, and no one outside Chainguard can instruct it or see its raw output. Every deliverable is reviewed and approved by a named Chainguard engineer before it reaches you. You'll be able to find and purchase Chainguard Harness directly through the Claude Marketplace, right alongside the other Claude-powered tools your team is already using. This makes high-end, AI-driven security accessible to vastly more teams.

We’re opening access to a limited group of organizations first. You can sign up for the beta waitlist today to save your spot.