What changed in Anthropic’s cyber access program?
Reuters reports that Anthropic is expanding its Cyber Verification Program so vetted cybersecurity professionals can use its most capable AI models with fewer safeguards. The new structure combines the earlier Project Glasswing and Cyber Verification Program tracks. It is tiered access for approved teams, not a public release without restrictions.
Glasswing partners reported at least 129,000 verified software vulnerabilities between April and July 2026. Anthropic’s open-source scans found another 5,500 through October. More than 33,000 of the reported findings were rated critical or high severity. Anthropic says its limited partner survey likely undercounts the impact; its estimate that the true impact could be at least five times higher is a company claim, not an independently verified total.
How do the three tiers work?
The Defense tier supports work such as incident response and malware analysis. Security teams, critical-infrastructure operators, open-source maintainers and researchers with a history of reporting vulnerabilities can apply. The Red Team tier adds authorized penetration testing and red-teaming, but only organizations can apply.
The Specialized tier has the fewest restrictions. It is reserved for a small group authorized to test safety-critical systems, including power grids, flight systems and interbank transfers. Reuters reports that Anthropic vets members of this tier together with the U.S. government. All three tiers include access to Claude Opus 5.5, Sonnet 5.5, Mythos 5.1 and future models, with differing verification and security controls.
Why it matters
AI systems able to uncover serious weaknesses could help defenders fix software before attackers exploit it. The same capabilities create misuse risks. The key questions are whether access controls hold, findings get verified and fixed, and expanded testing measurably improves security.
Watch, listen and read the source
Watch the Short above, listen to The Daily AI Chat on Spotify, or open the YouTube Short directly. This is a discussion of reporting and company claims, not a security audit or guidance for exploiting systems.
Source: Reuters, “Anthropic opens its most powerful AI models to more security teams”, October 6, 2026. Reporting by Anzar Mehraj and Jeffrey Dastin; editing by Leroy Leo.
