On October 7, Anthropic announced that it would expand access to its most advanced AI models, allowing verified organizations to utilize models such as Claude Opus 5.5, Claude Sonnet 5.5, and Claude Mythos 5.1 for cybersecurity testing. These tests will target critical infrastructure, including power grids, banking systems, and aviation operation systems, conducting 'high-risk offensive testing.' Anthropic stated that U.S. government agencies, financial institutions, and large software suppliers participating in the 'Project Glasswing' initiative will automatically gain access, while new organizations must undergo joint review by Anthropic and the U.S. government. Red teams will receive elevated privileges to conduct offensive security testing, but any actions that could cause physical harm or large-scale disruption are prohibited; certified defensive cybersecurity teams can leverage these models for tasks such as malware reverse engineering and incident response.