All articles tagged with #glasswing
Anthropic is expanding access to its cyber-capable AI model Mythos to about 150 organizations in more than 15 countries, including several EU members, via the Glasswing Project. ENISA has been invited to use the model, with access contingent on meeting strict security safeguards. The move aims to bolster cybersecurity defenses and signals ongoing expansion beyond the US to additional international partners.
Anthropic’s Project Glasswing, using Claude Mythos Preview, has identified more than 10,000 high- or critical-severity software vulnerabilities since launch, including 6,202 affecting over 1,000 open-source projects; 1,726 confirmed true positives and 1,094 high/critical. The effort has yielded 97 upstream patches and 88 advisories, with examples like WolfSSL CVE-2026-5194 (CVSS 9.1). The work underscores the gap between discovery and remediation, while defenders credit the tool for preventing fraud (a partner bank blocked a $1.5M wire transfer). Anthropic also launched a Cyber Verification Program for legitimate testing; public, unrestricted access to Mythos-like models remains unlikely amid safeguards. The episode amplifies calls to shorten patch cycles and harden defenses across software supply chains.
Anthropic revealed that Glasswing, powered by Claude Mythos Preview, autonomously identified over 10,000 high- and critical-severity zero-day vulnerabilities across critical software in its first month, with more than 50 tech partners including Microsoft, Apple, Google, and Cloudflare. Cloudflare alone found about 2,000 bugs (400 high/critical); Mozilla patched 271 Firefox vulnerabilities, while many disclosures have not been patched upstream, highlighting a severe patch-delivery bottleneck. The findings underscore the need for stronger defenses and faster triage, as Mythos-class models remain restricted to defenders, with Claude Security in public enterprise beta and industry groups rolling out supporting tools to cope with the AI-driven vulnerability deluge.