Anthropic has released an advisory regarding Project Glasswing, the AI initiative launched last month. This project is specifically designed to protect critical software systems from attacks by malicious AI models.
The introduction of Glasswing followed weeks after the revelation of Claude Mythos, a powerful new model developed by the company. Mythos demonstrated exceptional proficiency in identifying security vulnerabilities within code. Due to concerns about potential misuse if released broadly, Anthropic decided against a widespread public launch.
Instead, Anthropic chose to share Claude Mythos with approximately 50 key partners. This exclusive group includes some of the biggest names in the tech industry, such as AWS, Apple, Google, Microsoft, CrowdStrike, Nvidia, Broadcom, Cisco, and Palo Alto Networks.
Anthropic's early findings from Mythos have already yielded significant insights. Among these, the most striking are figures that illuminate the model's advanced capabilities in vulnerability detection.
