The most notable aspect of Claude’s AI cybersecurity model was discovering undetected vulnerabilities that remained 27 years after these systems were launched.