Learn about this first
Anthropic has launched OSS Scanner, an opt-in vulnerability scanning service for open-source projects. It uses its most capable models, including Claude Mythos, to regularly scan authorized projects for free. Reports are generated entirely by models, without human review or triage, and include proof of concept, vulnerability explanations, and remediation suggestions. Core maintainers submit an application PR to the OSS Scanner GitHub repository using a template. Priority is given to foundational projects with significant impacts on critical infrastructure and user security. Maintainers must provide a Dockerfile, and scans run in isolated virtual machines without network access.
According to MarkTechPost, the service reported 77 critical and 803 high-severity findings within its first 11 days. Of the 97 critical/high-severity findings verified by penetration testers, 88% met the disclosure threshold, and only 1 was a false positive. curl has since fixed CVE-2026-92392 based on these findings. The Decoder said the expected accuracy rate exceeds 90%, but reports may contain errors because they are not reviewed by humans. The Verge also noted that this approach may find vulnerabilities faster and more frequently, but could also produce erroneous or invalid reports.
AI-generated from reports · updated 22 days agoLatest developments
Anthropic launches free open-source vulnerability scanner OSS Scanner, reporting 77 critical and 803 high-severity findings in 11 days
Event progress1
Anthropic Launches OSS Scanner, an Open-Source Vulnerability Scanning Service
Anthropic: Research: Anthropic Launches OSS Scanner, a Free Open-Source Vulnerability Scanning Service