Anthropic Says Mythos Has Found More Than 10k Vulnerabilities (www.engadget.com)

Anthropic has announced that its cybersecurity initiative, Project Glasswing, has helped partners identify over 10,000 software vulnerabilities just one month post-launch. Utilizing the unreleased Claude Mythos Preview model, this initiative aims to leverage AI to mitigate threats from cyberattacks. Notable partners like Cloudflare and Mozilla reported significant bug-finding success, with Cloudflare identifying 2,000 vulnerabilities, while Mozilla found 271 vulnerabilities in Firefox—ten times more than with previous tools. Microsoft has also noted larger patch releases due to bugs uncovered through Mythos. The significance of this development lies in the innovative use of AI for cybersecurity, drastically improving vulnerability detection rates. Anthropic's work with partners, including major tech firms and the US government, reflects a proactive approach to security in an era where AI-related risks are increasingly pertinent. However, Anthropic has yet to release Mythos Preview publicly, citing the need for stronger safeguards to prevent misuse. The urgency for effective AI-driven security solutions is heightened by incidents reported in other systems, indicating a growing reliance on AI in safeguarding software integrity while the company anticipates potential profitability for the first time.