Anthropic revealed Claude Mythos Preview on April 7, positioning it as a general-purpose language model with standout performance in computer security. The model is not available through the standard API or consumer products. Instead, access is limited to Project Glasswing, a collaborative effort between Anthropic and a consortium of technology companies aimed at using Mythos to help secure critical infrastructure and software.
This is a notable strategic departure. Most AI labs race to ship new models publicly for revenue and developer mindshare. Anthropic is choosing restraint here, effectively saying the cybersecurity capabilities are powerful enough that controlled distribution matters more than broad access. For the industry, this sets a precedent for capability-gated releases.
If your product touches cybersecurity, watch for Project Glasswing partnership announcements - this could be a significant competitive advantage for those who get access early.
