Anthropic widens access to its most powerful AI models for vetted security teams

Anthropic is expanding a program that lets vetted cyber security professionals test its most powerful AI models with fewer safeguards. The move follows Project Glasswing, an initiative aimed at securing the world's most critical software. Partners found at least 129,000 verified vulnerabilities between April and July, and Anthropic's own open-source scanning found 5500 more between April and October. More than 33,000 have been rated critical or high severity. Anthropic says the figures are likely an undercount, and expects the true impact to be at least five times higher, because the data comes from a limited number of partners.

The revamped Cyber Verification Program merges Glasswing, which gave organisations access to Claude Mythos, with the original program that gave vetted teams reduced safeguards on Claude Opus and Sonnet. It has three tiers, each with its own verification requirements and security controls. The Defence tier covers work such as incident response and malware analysis, and is open to security teams, critical infrastructure operators, open-source maintainers and researchers with a record of reported vulnerabilities. The Red Team tier adds authorised penetration testing, but only organisations can apply.

The Specialised tier has the fewest restrictions and is reserved for a small group authorised to test safety-critical systems such as power grids, flight systems and interbank transfer infrastructure. Anthropic vets each member together with the US government, and existing Glasswing members will move into this tier. Anthropic's April unveiling of Claude Mythos Preview raised fears that AI could hack software before it had been secured.