Anthropic Launches Claude Mythos AI Model for Limited Corporate Use in Vulnerability Assessment

Anthropic announced the release of Claude Mythos last week, marking the latest addition to its family of AI models. This model is capable of writing code that can hack into critical systems, particularly those utilized by financial companies. Claude Mythos can analyze systems to identify and string together vulnerabilities, enhancing its effectiveness in cybersecurity tasks.

The release of Claude Mythos is limited to approximately 40 leading corporations. This initiative is part of an effort called Project Glasswing, aimed at allowing these corporations to assess the model's capabilities and prepare countermeasures against potential threats.

Jared Kaplan, co-founder and chief science officer at Anthropic, stated that Claude Mythos is the first model to demonstrate elite-level cybersecurity capabilities.

He noted that these capabilities arise not from a specific focus on cybersecurity during training but from the model's general intelligence and software proficiency. Kaplan emphasized that Claude Mythos excels at identifying vulnerabilities in software and understanding how to exploit them.

AI models are reportedly improving at a rate ten times faster than the historical trend of computer processing power doubling every 18 months, according to Kaplan.

This rapid advancement underscores the significance of Claude Mythos in the evolving landscape of artificial intelligence and cybersecurity.