Anthropic is moving to make its powerful and controversial Mythos-class AI model more widely available, aiming to bolster cybersecurity defenses globally.
Unveiled on April 7, 2026, Claude Mythos Preview can autonomously discover and exploit zero-day vulnerabilities in critical software like operating systems and browsers. It has already identified thousands of flaws, including a 27-year-old bug in OpenBSD and a 17-year-old issue in FreeBSD.
Instead of a public release, Anthropic launched Project Glasswing-a controlled initiative giving select partners access to use the technology defensively. Partners include AWS, Apple, Microsoft, Google, NVIDIA, CrowdStrike, and the Linux Foundation. Anthropic has committed $100 million in usage credits for participants.
However, on April 30, 2026, reports emerged that the White House opposes expanding access to roughly 70 additional organizations, citing the dual-use risk: a model that finds zero-days can also exploit them.
Anthropic argues that controlled proliferation is safer than keeping the technology locked away, as more defenders can patch vulnerabilities before bad actors find them. The White House appears to favor keeping access restricted for now.
For investors, the Project Glasswing partners hold a strategic edge. CrowdStrike could integrate Mythos-class capabilities into its enterprise products, creating a competitive moat. The inclusion of the Linux Foundation acknowledges the vulnerability of open-source projects to long-hidden bugs.
The $100 million investment in Glasswing serves as a proof-of-concept for responsible deployment. If partners show measurable improvements in vulnerability detection, it strengthens Anthropic's case for wider release. But political approval remains uncertain.