Anthropic has launched the Anthropic Project Glasswing, a major cross-industry initiative designed to protect critical software infrastructure from AI-powered cyber threats. The project brings together Amazon Web Services, Apple, Google, Microsoft and NVIDIA as the company pushes for faster cyber defence measures.
Anthropic said the initiative is powered by Claude Mythos Preview, a new frontier model that has demonstrated advanced ability to detect software vulnerabilities. According to the company, the model has already found several high-severity zero-day flaws across major operating systems and web browsers.
Anthropic said it created Project Glasswing because AI models now have coding abilities strong enough to outperform nearly all but the most skilled humans at finding and exploiting software weaknesses.
The company argued that this shift creates both a warning and an opportunity. On the one hand, advanced AI can expose dangerous flaws more quickly. On the other hand, defenders can use those same capabilities to secure essential systems before attackers move first.
A first look at Claude Mythos Preview, the model initially described in a leaked Anthropic draft as "by far the most powerful AI model we've ever developed."
So powerful, it's not getting released to the public.
The model will power Project Glasswing, an initiative with 12… pic.twitter.com/YPwqlMQrtX
— The Rundown AI (@TheRundownAI) April 7, 2026Claude Mythos Preview Finds Long-Standing Vulnerabilities
Anthropic said Mythos Preview independently discovered a flaw in OpenBSD, a 27-year-old piece of software. It also identified a bug in FFmpeg that had remained undetected for 16 years despite five million automated attempts.
In addition, the company said the model combined several Linux kernel vulnerabilities to take control of a system. According to Anthropic, these findings show how quickly AI-driven vulnerability research is advancing.
To support the rollout, Anthropic said it will allocate $100 million in usage credits tied to Mythos Preview. The company also plans to donate $4 million directly to open-source security organisations.
Read: OpenAI Acquires Astral for Codex in Push Against Anthropic
However, Anthropic said it does not plan to release the software publicly. That decision suggests the company wants to limit broader access while using the model in controlled security work.
Anthropic warned that defending the world’s cyber infrastructure could take years, even as frontier AI capabilities may improve substantially within months.
The company said cyber defenders must act now to avoid falling behind. That message underscores the urgency of Project Glasswing and the broader race to secure critical systems in the AI era.
