Anthropic Mythos Preview exploits thousands of bugs

This is less a consumer launch than a proof point that frontier coding and agentic models now cross into offensive-security territory by default.

• –Anthropic is limiting access through Project Glasswing, with launch partners using the model for defensive work rather than a broad public rollout.
• –The real implication is dual-use: the same capability that helps defenders patch systems faster can also compress attacker timelines for exploit discovery.
• –The 27-year-old bug detail matters because it shows the model is useful on obscure, long-lived flaws, not just fresh CVEs.
• –For developers, this raises the pressure on secure-by-default code, tighter red-teaming, and stricter controls around releasing highly capable models.
• –It also signals that “too dangerous to ship broadly” is becoming a real product category, not just a safety slogan.