Anthropic has decided to withhold public access to its Mythos AI model after tests conducted by JPMorgan, Apple, Google and others. Mythos Preview has demonstrated the ability to uncover complex software flaws, a capability that raises serious cybersecurity concerns about how such tools are used and controlled.
To address these risks, the company is keeping Mythos behind a controlled access framework and moving toward a private patch collaboration through Project Glasswing, which partners with major tech players to identify and patch vulnerabilities rather than release broadly.
Glasswing brings together Apple, Google and Microsoft to leverage Mythos' insights for strengthening cybersecurity worldwide, aiming to patch critical vulnerabilities that could otherwise be exploited at scale.
Industry observers say the decision reflects a broader shift in AI governance—balancing breakthrough capabilities with safeguards, and signaling that public access may come later only after robust risk controls are in place.
