Anthropic has unveiled Project Glasswing, an AI model that identifies security issues across major operating systems and web browsers. The company's goal? To provide its 'defensive security' partners with early warnings against cyber threats, all without human intervention.

The new general-purpose model, Claude Mythos Preview, boasts a ‘head start’ advantage for defenders, having flagged thousands of high-severity vulnerabilities in recent weeks. Yet, as it autonomously discovers and exploits weaknesses, concerns over potential misuse linger.

While Anthropic is currently offering the tool to select partners like JPMorgan Chase and the Linux Foundation, there's talk of turning it into a paid service if it proves effective enough. Meanwhile, the company remains in discussions with US government officials about its capabilities.

Dianne Penn from Anthropic stated that they are 'taking steps' to solidify their processes following a data leak attributed to human error. However, these protective measures may not be sufficient; as AI evolves, so do the risks it presents.