Microsoft Tests Mythos to Identify and Mitigate Vulnerabilities

Microsoft announced on April 22, 2026, that it is testing Anthropic’s Claude Mythos Preview via Project Glasswing with partners to identify and mitigate vulnerabilities earlier. Using its CTI-REALM benchmark, Microsoft found Mythos delivered “substantial improvements” versus prior models and plans to embed Mythos and other advanced AI models into its Security Development Lifecycle to surface issues, guide customers on exposure reduction, and build defensive solutions. Project Glasswing includes major tech and cybersecurity players (Amazon, Apple, Palo Alto Networks, Google, Nvidia, Crowdstrike), potentially accelerating enterprise adoption of AI-powered security tools. Anthropic is probing reported unauthorized access to Mythos, a development that could temper near-term adoption or raise governance concerns. Market impact: the initiative reinforces Microsoft’s security positioning and its AI integration strategy, which could be viewed positively for MSFT and partner vendors providing AI-security solutions, while the access investigation is a short-term risk to sentiment around model rollouts.