OpenAI and Anthropic Strengthen Access Control for High-Performance AI

OpenAI and Anthropic are narrowing access to high-performance AI models, focusing on identity verification and approval processes. OpenAI announced in an official document dated May 7, 2026, that it will operate GPT-5.5 and GPT-5.5-Cyber under a 'trust-based cyber access' system. For verified defensive purpose users, restrictions on tasks such as vulnerability identification and malware analysis will be reduced, while credential theft and malware distribution will be blocked. Anthropic provides Claude Mythos 5 only to a select group of verified partners, explaining that while this model performs strongly in cybersecurity and biological research, it also has potential for misuse. The trust-based access approach varies the scope of permissions based on user identity and intended use, requiring AI companies to distinguish between legitimate security research and dangerous attack requests. This trend indicates a shift in the competition among AI companies from performance disclosure to access control and safety mechanisms. On June 12, Anthropic temporarily suspended access to Claude Fable 5 and Mythos 5 due to U.S. government export controls, restoring some access on July 1. Access control presents both burdens and opportunities for the AI infrastructure market, and access rights to high-performance models can impact the speed of vulnerability verification and response.