Anthropic expands Claude Mythos Preview access — security or marketing gimmick?
Too dangerous to release? Anthropic opens its 'catastrophic' cyber model to 150 orgs.
Anthropic is expanding access to Claude Mythos Preview, a model designed to find critical security vulnerabilities, through its Project Glasswing initiative. The model is now available to 150 new organizations across 15 countries, including those in power, water, healthcare, communications, and hardware. While Mythos excels at exploit chain construction, it has inconsistent safeguards and struggles to write patches without breaking code. Critics argue the limited rollout is more marketing than true safety precaution.
- Claude Mythos Preview is now available to 150 organizations across 15 countries in critical infrastructure sectors.
- The model excels at exploit chain construction but has inconsistent safeguards and often breaks code when patching.
- Critics and security experts argue the limited rollout is a marketing gimmick, not a genuine safety measure, as open-source models find similar vulnerabilities.
Why It Matters
Highlights tension between responsible AI deployment and marketing hype in cybersecurity—professionals must watch for real value vs. exclusivity.