Anthropic launched Project Glasswing, a closed cybersecurity initiative deploying its new frontier model Mythos with 12 major partners to find critical software vulnerabilities.
Anthropic released a limited preview of Mythos, a new frontier general-purpose model with strong agentic coding and reasoning capabilities, exclusively for cybersecurity defensive work. The initiative, Project Glasswing, includes 12 partner organizations — among them Amazon, Apple, Microsoft, Cisco, CrowdStrike, and the Linux Foundation — with 40 organizations total gaining preview access. Anthropic claims Mythos identified thousands of zero-day vulnerabilities, many of them critical and some one to two decades old, within just a few weeks of deployment. The model is not yet publicly available; partners will share learnings with the broader tech industry.
Mythos is performing agentic code vulnerability scanning at a scale and depth no human security team can match — finding critical zero-days that sat undetected for 20 years in widely-used software. This signals that frontier model agentic coding capability has crossed a meaningful threshold: it's not just autocomplete, it's autonomous vulnerability discovery. Developers shipping open-source or widely-deployed software should treat this as a signal that AI-assisted code auditing is no longer a nice-to-have.
Run Claude's existing API with a code analysis prompt against your most critical module this week — before Mythos-class capability reaches adversarial actors who will use it for exploitation, not defense.
Go to claude.ai and open a new conversation
Tags
Related
Also today
Signals by role
Also today
Tools mentioned