.svg){kind=spacer}
Details
- Anthropic introduces Project Glasswing, an initiative using its newest frontier model, Claude Mythos Preview, to detect software vulnerabilities surpassing most human experts.
- Partners include Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks to identify and fix flaws in essential systems.
- Mythos Preview has discovered thousands of high-severity vulnerabilities in every major operating system and web browser.
- Anthropic commits up to $100M in usage credits for partners and over 40 organizations maintaining critical software, including open-source projects, and will report findings.
- Model not for general release; focus on deploying safeguards first, with testing on upcoming Claude Opus model.
- Detailed technical report and Mythos Preview system card available via shared links.
Impact
Project Glasswing positions Anthropic ahead in defensive AI cybersecurity by granting early access to Mythos Preview, a model outperforming prior Claude Opus 4.6 in vulnerability detection, as seen in its prior discoveries of 22 Firefox flaws and Linux kernel issues. This pressures rivals like OpenAI and Google to accelerate similar tools amid rising AI-driven exploits, while giving defenders a head start to patch critical systems before widespread model availability narrows the gap. It aligns with industry calls for proactive hardening against agentic AI threats, potentially shifting vulnerability timelines from weeks to hours.
