Anthropic announced Claude Mythos Preview and launched Project Glasswing, a coordinated defense initiative with major partners including AWS, Google Cloud, Microsoft, NVIDIA, Cisco, and others.
The claim is bold: Mythos Preview can autonomously find and exploit serious vulnerabilities across major operating systems and browsers, including long-standing bugs that survived years of conventional review.
Instead of broad release, access is being gated to selected organizations focused on defensive security work. Anthropic says it is pairing the launch with major usage credits and open-source security funding.
Why this matters
- The AI security race just moved from theory to deployment. Vulnerability discovery and exploit generation are now scaling faster than human-only teams can match.
- Patch velocity becomes strategic. If capability diffusion accelerates, secure-by-default engineering and shorter remediation loops become non-optional.
- Winners will be operational, not rhetorical. The leading organizations in this phase will have tighter secure development pipelines, faster incident response, and stronger AI-native defense stacks.
In short: this is less about a single model launch and more about a new baseline for cyber offense/defense speed. The companies that adapt fastest at the systems level will matter more than the companies that simply claim better model intelligence.