Anthropic opens three-tier cyber access programme to verified security teams
Anthropic has merged Project Glasswing and its Cyber Verification Program into three access tiers for approved cybersecurity professionals, covering defensive work, authorised red-teaming and testing safety-critical systems. The company says partners identified at least 129,000 verified software vulnerabilities from April to July, including more than 33,000 rated critical or high severity; it calls the figures an undercount.
Bottom line — Anthropic’s three tiers set different limits on access to Claude Opus 5.5, Sonnet 5.5 and Mythos 5.1.
Go deeper 7
-
Defense Access covers incident response, malware analysis and vulnerability validation; Anthropic says it aims to respond to applicants within days.
-
Red Team Access permits authorised penetration testing, but is limited to organisations and takes weeks to review, according to Anthropic.
-
Specialized Access has the fewest cyber safeguards and is reserved for vetted organisations testing systems including power grids, aviation, telecoms and interbank transfers.
-
Anthropic says Specialized Access applicants are vetted alongside the US government; existing Project Glasswing members move into the tier without reapproval for current models.
-
Anthropic’s own open-source scanning added 5,500 vulnerabilities between April and October, according to the company.
-
More than 33,000 vulnerabilities were rated critical or high; Anthropic says the total impact may be at least five times higher, based on a limited subset of partner reports.
-
The Daily Star reports that data retention is required to monitor for misuse, while Anthropic says a system allowing organisations to store data in their own cloud infrastructure is due this autumn.