Schneier on Security

Claude Opus 4.6 demonstrates significant advancement in automated vulnerability discovery, finding high-severity zero-day vulnerabilities in well-tested codebases without specialized tooling. Unlike traditional fuzzers that use random inputs, the model reasons about code like a human researcher—analyzing past fixes, recognizing problematic patterns, and understanding logic to craft precise exploits. It successfully discovered vulnerabilities that had remained undetected for decades in projects with millions of hours of fuzzing.