On Tuesday, the AI startup Anthropic released a preview of its new model, Mythos. Designed for cybersecurity work, this powerful tool will be used by partner organizations including Amazon and Apple in their defensive security efforts.

The model has already identified thousands of zero-day vulnerabilities, some dating back decades. While it was not specifically trained for cybersecurity, Anthropic claims that Mythos can scan both first-party and open-source software systems for code weaknesses. The company is also engaging with federal officials about the use of this powerful AI.

However, Anthropic's journey to innovation has been rocky. Earlier this year, a draft blog about their new model was leaked due to human error, revealing it as a far more potent tool than previously known ones like Opus. The leak also hinted at potential cybersecurity risks if the technology were weaponized.

Project Glasswing, the initiative involving 12 partner organizations and 40 total participants, aims not only for defensive security but also to share insights with the broader tech community. Despite this, Anthropic remains cautious about making Mythos generally available until they can ensure its responsible use.