Anthropic, a San Francisco-based AI company, has introduced Project Glasswing, a cybersecurity initiative featuring the AI model Claude Mythos Preview. The project aims to identify and patch software vulnerabilities in critical infrastructure before they can be exploited by malicious actors. Anthropic has partnered with major tech and finance companies, including Amazon Web Services, Apple, Microsoft, and Nvidia, to enhance cybersecurity defenses.
While Anthropic’s revenue has grown significantly, reaching over $30 billion annually, the company made the decision to keep the Claude Mythos Preview model restricted due to its potential cybersecurity risks. The model autonomously discovered numerous zero-day vulnerabilities in essential software, prompting Anthropic to withhold its release to provide defenders with a strategic advantage.
Project Glasswing’s approach to disclosing vulnerabilities involves a meticulous triage process to prevent overwhelming open-source maintainers. Anthropic plans to collaborate with maintainers to develop and validate patches, ensuring the timely and secure deployment of fixes.
The initiative has received support from industry leaders like CrowdStrike, Microsoft, and the Linux Foundation, who have tested the Mythos Preview model against their infrastructure and cited substantial improvements in security measures. Anthropic’s partnerships and financial commitments underscore the company’s dedication to advancing cybersecurity capabilities.
Source: VentureBeat