Anthropic has announced Project Glasswing, restricting access to their new Claude Mythos model to a select group of security research partners rather than releasing it publicly. The model has demonstrated unprecedented capability in finding vulnerabilities — including a 27-year-old OpenBSD kernel bug and Linux privilege escalation flaws — by chaining multiple vulnerabilities into sophisticated exploits. Security professionals like Greg Kroah-Hartman and Daniel Stenberg confirm a recent shift from AI-generated security noise to genuinely high-quality AI-assisted vulnerability reports. Anthropic is backing the initiative with $100M in usage credits and $4M in donations to open-source security organizations, with partners including AWS, Apple, Microsoft, Google, and the Linux Foundation. The author endorses the restricted rollout as a reasonable precaution given the real and credible security risks.
Sort: