Anthropic drops restricted Claude Mythos cybersecurity model

Claude Mythos marks the transition of LLMs from coding assistants to autonomous security researchers capable of unearthing decades-old vulnerabilities.

• –Achieves a record-breaking 93.9% on SWE-bench Verified, nearing human-level software engineering autonomy.
• –Autonomously discovered thousands of zero-day vulnerabilities in major operating systems, including a 27-year-old OpenBSD bug.
• –Restricted access strategy underscores the growing "dual-use" risk where frontier models become viable offensive weapons.
• –High-cost pricing of $125 per million output tokens positions it as a premium, specialized tool for mission-critical infrastructure.
• –Strategic partnership with AWS, Google, and NVIDIA suggests a coordinated effort to secure global software supply chains before malicious actors weaponize similar capabilities.