ANTHROPIC will expand access to its latest AI models, allowing vetted institutions to use Claude OPUS 5.5, Claude Sonnet 5.5 and Claude MYTHOS 5.1 for cybersecurity testing, including "high-risk offensive tests" targeting critical infrastructure such as the power grid, banks and aviation operations. Participants in its "Project Glasswing"—US government agencies, financial institutions and large software vendors—will receive access; new entrants require joint ANTHROPIC–US government review. Red teams will be granted elevated privileges for offensive testing, though actions likely to cause physical harm or mass destruction are prohibited. Certified defensive cybersecurity teams may use the models for malware reverse engineering and incident response. 06-OCT-2026 03:50 PM ET