Chainguard has released an AI-powered security tool that uses frontier models, including Anthropic's restricted Claude Mythos 5.1, to identify and repair exploitable weaknesses in application code that traditional scanners overlook, the company announced Wednesday. The system, called Harness, has uncovered serious vulnerabilities in codebases that had already gone through extensive security scanning, including software deployed in critical infrastructure. Harness leads a four-product launch alongside Workspaces, Checks, and expanded Libraries protections, pushing Chainguard's offerings beyond hardened container images and dependencies into agent execution environments and continuous integration verification.
Anthropic has limited Claude Mythos 5.1 access to organizations participating in Project Glasswing, leaving unclear which models Harness customers outside the program will use, according to the announcement. Developers can use Chainguard Harness without direct access to Mythos, but the company hasn't disclosed which models handle vulnerability discovery for those customers. The tool will also be available through Anthropic's Claude Marketplace, allowing teams to tap existing Anthropic spending commitments rather than secure a separate budget. The company hasn't revealed details about the affected projects, the flaws themselves, or their remediation status.
The launches arrive alongside The AI Exploit Gap, a Chainguard-commissioned Wakefield Research survey of 1,300 security leaders across nine markets. More than half said their organizations knowingly leave high or critical vulnerabilities unpatched at least sometimes, with the figure reaching 66% at companies with $1 billion or more in revenue, compared with 27% at companies below $250 million. "AI can find and chain flaws in minutes, while fixing them still moves at human speed and competes with product deadlines and uptime," Quincy Castro, Chainguard's chief information security officer, said. Castro added that patching faster isn't viable for most organizations, and it will consume increasing amounts of resources just to keep pace.
The security harness addresses what the company describes as a growing asymmetry: AI can discover and chain vulnerabilities in minutes, but remediation still proceeds at human velocity and must compete with product deadlines and system uptime requirements. Each developer and coding agent receives a hardware-isolated microVM with network policies that restrict access to internal systems, keeping credentials outside the agent's environment to limit what a compromised agent or malicious dependency can reach. Chainguard Checks lets coding agents run continuous integration checks directly during development and receive signed results in seconds, allowing agents to run tests, inspect failures, and revise code before opening a pull request while avoiding the CI queues that become harder to manage as vendors turn coding agents into shared team infrastructure running multiple sessions in parallel.
Chainguard Libraries is adding secure package endpoints for Go and .NET, along with CVE remediation for JavaScript and Go, with Java and Python support expanding to cover newly disclosed vulnerabilities discovered through AI-assisted research, including flaws in dependencies that coding agents may have already pulled into a project. The changes build on Chainguard Repository, a policy-enforced endpoint for open source dependencies, and its partnership with Cursor, allowing coding agents to resolve packages through Chainguard's catalog instead of pulling directly from public registries. Checks is available now in beta, with signed activity logs recording what each agent did so teams can maintain an audit trail across multiple autonomous sessions. The inclusion of restricted frontier models in commercial security tools suggests the boundary between experimental AI capabilities and production infrastructure is dissolving faster than procurement cycles can adapt, forcing security teams to evaluate black-box systems whose internals they can't inspect. Organizations that lack Glasswing access will face a trust calculus with no technical fallback, betting operational security on model performance they can neither verify nor replicate independently.