Anthropic has developed a new AI model called Claude Mythos, described as extremely powerful and capable of finding security vulnerabilities in software. Due to its potential for misuse by bad actors, Anthropic is not releasing it publicly. Instead, through a program called Project Glasswing, access is being granted exclusively to cybersecurity specialists at select companies. The goal is to help these organizations proactively discover and patch vulnerabilities before similarly powerful models become widely available. As a demonstration of its capabilities, Mythos preview reportedly uncovered a 27-year-old vulnerability in OpenBSD.
•1m watch time
Sort: