What Anthropic announced
Anthropic is extending Project Glasswing, a program that gives partners access to the Claude Mythos Preview model to find and fix vulnerabilities in their own systems, to roughly 150 more organizations in over 15 countries. Anthropic describes Mythos as a frontier model capable of finding and exploiting software vulnerabilities at a level beyond most skilled human testers. The program started with about 50 partners in April, including Apple, Nvidia, Microsoft, CrowdStrike, and Palo Alto Networks.
Reported results so far
Anthropic says participating partners have identified more than 10,000 high or critical severity flaws since launch. Reported use cases include local vulnerability detection, black-box testing of binaries, endpoint hardening, penetration testing, threat detection, and translating code into memory-safe languages.
Who gets access now
According to CNBC and Cybersecurity Dive, the new cohort emphasizes critical-infrastructure operators in sectors that were underrepresented at launch, such as power, water, healthcare, communications, and hardware. CNBC reports that new partners must satisfy unspecified security requirements before they are granted access.
Analysis (generic industry view)
- •Models that can both discover and exploit vulnerabilities are inherently dual-use, so vetted, staged rollouts are the common industry approach to balancing defensive value against misuse risk.
- •Broadening the tester pool typically surfaces more environment-specific and rare exploitation paths than a single in-house red team encounters.
- •The headline flaw count is self-reported program data; independent disclosure timelines and remediation rates are the more durable signals to watch.
Key Points
- 1Anthropic says Project Glasswing partners have surfaced more than 10,000 high or critical severity flaws since the April launch (Anthropic, CNBC).
- 2The expansion adds critical-infrastructure sectors (power, water, healthcare, communications, hardware) across 15-plus countries, with security vetting required before access.
- 3Industry pattern: frontier models that can both find and exploit vulnerabilities are dual-use, so staged, vetted access is the common way labs balance hardening against misuse risk.
Scoring Rationale
A frontier model that can find and exploit software vulnerabilities being extended to roughly 150 critical-infrastructure partners across 15-plus countries, with more than 10,000 reported flaws, is a significant AI-security development with clear dual-use stakes. It sits in the notable-to-major band but stops short of industry-shaking, since access stays gated and results are self-reported. Raised slightly from 6.8 to reflect the critical-infrastructure scope.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
