Home / Technology / AI Giants Open Up Powerful Models for Security Tests
AI Giants Open Up Powerful Models for Security Tests
7 Oct
Summary
- Cybersecurity professionals tested powerful AI models with fewer safeguards.
- Project Glasswing found over 100,000 software vulnerabilities this year.
- The revamped Cyber Verification Program has three tiers with different controls.
Anthropic has broadened its initiative that grants cybersecurity professionals access to test its most powerful AI models with fewer restrictions. Project Glasswing, launched to secure critical software globally, identified at least 129,000 verified software vulnerabilities between April and July. Anthropic's own scanning efforts added another 5,500 vulnerabilities between April and October.
The newly announced Cyber Verification Program (CVP) consolidates two previous programs. The Defense tier focuses on incident response and malware analysis, available to security teams and researchers. The Red Team tier permits authorized penetration testing, exclusively for organizations. The Specialized tier, with the fewest restrictions, is for a select group testing safety-critical systems, with members vetted alongside the U.S. government.
