Today, Anthropic unveiled "Project Glasswing," an initiative to protect critical software from the very AI capabilities it is racing to develop.
The centerpiece is Claude Mythos Preview — their latest unreleased frontier AI model. Anthropic admits it represents a “particularly large step up” in coding and reasoning, enabling it to uncover thousands of high-severity zero-day vulnerabilities in every major operating system, web browser, and other essential software. Many of these flaws had survived years of human scrutiny.
Instead of releasing Mythos Preview to the public, Anthropic is granting early, controlled access only to who they call the “good guys” in cybersecurity. Partners include Amazon Web Services, Apple, Google, Microsoft, NVIDIA, CrowdStrike, Palo Alto Networks, JPMorgan Chase, the Linux Foundation, and over 40 other organizations.
These "defenders" will use the model to scan and patch vulnerabilities before similar AI tools fall into the hands of malicious actors.
Anthropic is committing up to $100 million in usage credits and $4 million in donations to open-source security efforts.
While the move sounds responsible on the surface, it raises perhaps uncomfortable questions. AI labs keep building ever-more-powerful systems whose side effects include turning software security upside down — then position themselves as the "heroes" offering controlled fixes. One can’t help but wonder: Are we merely racing to clean up messes that frontier AI is creating faster than we can address them?
Project Glasswing may buy us some time, but it also highlights how deeply the world is becoming dependent on these rapidly advancing systems.