Stories
Anthropic Expands Access to Its Most Powerful AI Models for Security Teams
F
Firstpost•07-10-2026, 10:24
Anthropic Expands Access to Its Most Powerful AI Models for Security Teams
- •Anthropic is expanding its security program after Project Glasswing helped uncover over 100,000 software vulnerabilities this year.
- •Glasswing partners found at least 129,000 verified vulnerabilities between April and July, with over 33,000 rated critical or high severity.
- •The revamped Cyber Verification Program combines previous security initiatives and provides vetted professionals access to advanced models like Claude Mythos and Sonnet.
- •The new program features three tiers—Defense, Red Team, and Specialized—with varying verification requirements and security controls depending on the intended use case.
- •The most restrictive Specialized tier is reserved for a small group of organizations testing safety-critical systems like power grids and flight infrastructure, vetted alongside the US government.
Key Questions Answered
The revamped Cyber Verification Program (CVP) combines Anthropic's previous Project Glasswing and original CVP initiatives, providing vetted cybersecurity professionals access to powerful models like Claude Opus 5.5, Sonnet 5.5, and Mythos 5.1 under three distinct security tiers.
The revamped program features the Defense tier for incident response and malware analysis, the Red Team tier for authorized penetration testing, and the Specialized tier, which has the fewest restrictions for testing safety-critical systems like power grids.
Partners under the Project Glasswing initiative found at least 129,000 verified software vulnerabilities between April and July, with over 33,000 rated as critical or high severity.
