Anthropic's New AI Model Enters the Cybersecurity Arena with Elite Partners

Anthropic is introducing a new, private AI model through a coalition of major technology and financial firms. Dubbed Project Glasswing, the initiative aims to provide large corporations and government entities with a tool to autonomously identify and help remediate critical security flaws in their software systems.

The core offering is Claude Mythos Preview, a general-purpose model Anthropic does not intend for public release. According to Newton Cheng, who leads cybersecurity efforts for Anthropic's frontier red team, the goal is to offer defenders an advantage. Partners, including Nvidia, Google, Amazon, Microsoft, Apple, JPMorgan Chase, and several cybersecurity giants, will use the model to scan their infrastructure. Access is tightly controlled to prevent malicious use.

While not specifically trained for security, Anthropic credits the model's advanced coding and reasoning abilities for its performance. The company reports that in recent tests, Mythos Preview autonomously discovered thousands of high-severity vulnerabilities across every major operating system and web browser, even developing related exploits without human guidance.

The model's existence became known last month due to an internal data leak, which Anthropic attributes to procedural error. For now, the company is subsidizing access for its partners with up to $100 million in usage credits. If successful, the program could evolve into a commercial service.

Anthropic also confirmed it has briefed senior U.S. government officials on the model's capabilities, indicating ongoing dialogue about its potential defensive and offensive applications in national cybersecurity.

Source: The Verge