Anthropic expands its Cyber Verification Program by integrating Project Glasswing and offering three tiers, all with access to its most capable Claude models
We're launching a new, expanded version of our Cyber Verification Program (CVP), which makes advanced cyber capabilities …
Context & Ripple Effects
Project Glasswing began as a restricted effort to put Claude Mythos Preview in the hands of more than 40 organizations maintaining critical software, backed by limited access for critical-software maintainers and up to $100 million in usage credits. Anthropic later said the initiative had helped identify more than 10,000 high- or critical-severity vulnerabilities. That reported vulnerability volume gives the consolidation into CVP operational weight beyond a program rebrand.
The new structure turns an initiative with a defined cybersecurity mission into a tiered access program, making verification and access levels part of Anthropic's route for deploying its latest models to cyber organizations.
First-order effects
- Anthropic replaces separate CVP and Project Glasswing tracks with one three-tier program, giving participating cyber organizations a single framework for access to its latest models.
- Organizations that entered through Glasswing move from a bounded vulnerability-focused initiative to a broader CVP structure with differentiated access levels.
Second-order effects
- Anthropic can use CVP tiers to align model availability with the level of verification applied to cyber organizations, rather than treating the Glasswing cohort as a separate access channel.
- The earlier credit-backed Glasswing effort becomes part of a repeatable program model, shifting the emphasis from a launch initiative to ongoing participant management.
Third-order effects
- If this structure is sustained, advanced cyber-model distribution will increasingly be organized around verified, tiered programs instead of broad model releases or isolated security pilots.
- The move points to capability tiering as a deployment mechanism for high-risk AI uses: model providers can expand access while preserving distinct controls for different user groups.
The trend: AI providers are turning restricted cybersecurity model access into tiered verification programs that connect capability access to operational controls.