Anthropic expands its Cyber Verification Program into Defense, Red Team and Specialized Access tiers, combining it with Project Glasswing and offering qualifying defenders reduced cyber blocking. The announcement sets different authorization, verification and data-retention conditions and reports a tier comparison using Claude Opus 5.5. Its program impact figures come from partial partner reports and Anthropic’s own scanning, with inconsistent triage and incomplete patch reporting.
Why it matters
It documents how access to stronger cyber capabilities is governed, and how blocking differs between ordinary defensive and authorized adversarial work.
Key facts
As stated in the sources, with where to find them.
- Defense Access covers defensive work and can include individual vulnerability researchers. Red Team Access adds authorized adversarial testing and is limited to organizations, with blocks retained for actions that could cause physical harm or mass disruption. Specialized Access has fewer cyber blocks for a limited set of organizations reviewed in depth with the US government.New access tiers
- On ten CyScenarioBench challenges with five attempts per tier, the generally available configuration was blocked on the first prompt for every task. Defense Access blocked 46 of 50 trials at some point; four succeeded. Red Team Access had no blocks and completed 34 of 50 tasks. These are vendor-run trials of Opus 5.5, not general safety guarantees.Testing the efficacy of our tiers
- Data retention is generally required for misuse monitoring. Anthropic describes exceptions for organizations with existing qualifying zero-data-retention access and a planned Enterprise Frontier Safeguards service; the planned service is not recorded as already available.New access tiers
- Anthropic reports at least 129,000 verified vulnerabilities from Glasswing partners between April and July and 5,500 more from its open-source scans between April and October. Anthropic says more than 33,000 of the verified vulnerabilities across those efforts were rated critical or high severity. The figures rely on partial data from 33 partner reports and open-source partnerships, heterogeneous triage, and patch numbers disclosed by fewer than half of partners. They do not establish a comparable patch rate.Giving defenders the advantage
Findings that cite this record
Key questions this bears on
- Is AI shifting the balance between finding and fixing vulnerabilities?Discovery is ahead. Verification, triage, coordinated disclosure and safe repair remain bottlenecks between AI findings and deployed fixes.
Sources
Related records
Sep 22, 2026
Sep 29, 2026
May 22, 2026
Oct 21, 2026
Oct 9, 2026
Jun 30, 2026