Anthropic Unveils Claude Mythos Model
On May 28 at 21:40, Anthropic announced preparations for the public launch of its Claude Mythos artificial intelligence model, which has shown exceptional skill in generating exploits and uncovering vulnerabilities. During testing, this model successfully produced working exploits in 72.4% of cases, highlighting its potential impact on cybersecurity.
The model has already scanned over a thousand open-source projects, identifying 23,000 errors—6,000 of which were classified as critical. Notably, Claude Mythos discovered a severe vulnerability in the wolfSSL cryptographic library that could have allowed attackers to forge certificates and impersonate legitimate websites. This flaw has since been patched.
Plans for Broader Access
Details on expanding access to Claude Mythos are outlined in an update to the Project Glasswing initiative. However, Anthropic representatives caution that no organization currently has adequate safety measures in place, meaning a full public release of Mythos-class models will only occur once proper safeguards against misuse are established.
Anthropic's official blog states that 'Mythos-level models will become widely available within the next 6 to 12 months.'
Claude Mythos was first introduced in April, and experts are now focused on ensuring the technology's security before it is made available for widespread use.
The launch of Claude Mythos underscores the growing role of artificial intelligence in cybersecurity. Its ability to detect vulnerabilities in widely used libraries like wolfSSL demonstrates the value of automated solutions for protecting information systems. Yet, before the technology goes public, robust security measures must be implemented to prevent potential abuse—a critical step given the escalating cyber threats.