Anthropic has presented Project Glasswing, a cybersecurity initiative that uses its new AI model, Claude Mythos. The project focuses on the proactive identification of security flaws. In its initial phase, the system has already detected thousands of zero-day vulnerabilities in infrastructures of major technology firms, including Amazon Web Services, Apple, and Cisco.
Architecture and approach of an AI model for code auditing 🔍
Claude Mythos operates through large-scale code analysis and simulation of possible attack vectors. It does not limit itself to searching for known patterns, but models the behavior of complex systems to foresee logical and configuration flaws. The process involves structural scanning and causal reasoning, identifying points where operation sequences could be exploited. The findings are reported confidentially to the affected companies.
The AI interns already have work for the next ten years ☕
With thousands of zero-day flaws discovered in just the first phase, the patch teams of the big tech companies must be delighted. Imagine the inbox full of vulnerability reports before taking the first coffee. It's like they hired a relentless auditor who never sleeps and whose only hobby is pointing out every crack in the wall. A job boost for the cybersecurity sector, without a doubt.