Back to the stories

Anthropic launches opt-in OSS Scanner vulnerability-finding service

Score 9.0,

AI cybersecurity

Anthropic is offering open-source projects an opt-in service that finds software vulnerabilities using its Claude models.

Called OSS Scanner, the service scans public codebases to surface security issues. Anthropic says OSS Scanner is informed by Project Glasswing, where it used Claude to locate chained vulnerabilities, and by its Frontier Red Team, which stress-tests AI systems for cyber, national-security and autonomous-system risks.

Why this matters: open-source projects often rely on volunteers or external audits to find bugs. An automated, AI-assisted scanner can flag problems earlier and across far larger codebases, which changes who discovers and fixes vulnerabilities.

How it works: think of OSS Scanner as a tireless security consultant that reads huge codebases and highlights risky patterns for humans to inspect. Claude proposes likely flaws; the Frontier Red Team's methods shape how those probes are run and evaluated. Because projects must opt in, maintainers control when scans run and who sees results.

What changes now: open-source maintainers gain a new, optional tool to spot flaws earlier. Organizations that depend on open-source get another layer of checking before they deploy code.

This is not a magic fix. Findings still need expert review and careful disclosure, and widespread use will depend on how accurate scans prove and how well maintainers handle the reports.

The real question is adoption: will maintainers trust AI-driven scans, and can Anthropic demonstrate accuracy and responsible handling? Watch for a GA rollout, public benchmarks, and any regulatory responses.