OpenAI and Anthropic Limit Latest AI Models to US Government-Approved 'Trusted Partners'
Tags AI · Infrastructure · Policy
OpenAI restricted release of its GPT-5.6 model and Anthropic limited access to Claude Opus 4.8 at the request of the US government, requiring customers to be vetted 'trusted partners' during a cybersecurity review. The restrictions, first reported by CNBC and the Washington Post on June 26, apply to the most powerful new model releases from both frontier AI labs. OpenAI confirmed the limits were imposed under pressure from the Trump administration, while Anthropic said it was working with the government on a security review. The policy creates a two-tier access model where only US-government-approved organizations can obtain cutting-edge capabilities.
Technical significance
This creates the first formal government-gated access tier for frontier AI models, establishing a precedent where model developers act as enforcement arms of export-control policy. For developers outside the vetted partner program, this means delayed or denied access to the latest capabilities, potentially widening the gap between US-aligned and non-aligned AI ecosystems.