The US government now runs the release schedule for the world’s most advanced artificial intelligence. For the first time, the executive branch has placed national security holds on commercial models, dictating who can use OpenAI’s GPT-5.6 and Anthropic’s Mythos 5.
Commerce Secretary Howard Lutnick is the gatekeeper. He reauthorized access to Mythos 5 for about 100 “trusted partners” after Anthropic addressed model risks, and OpenAI restricted GPT-5.6 to a similar-sized, government-approved list. On Hard Fork, Casey Newton and Kevin Roose argued this creates a de facto licensing regime run on opaque administrative whims, not technical standards or law.
“The shift to a 'default no' environment means labs must assume their next breakthrough will be mothballed by the federal government until further notice.”
- Hard Fork
The process is discretionary and politically fraught. OpenAI’s Sam Altman publicly backed safety reviews but called the current process “not quite the process that we think is optimal.” Even significant political donations haven’t guaranteed access; OpenAI’s Greg Brockman donated $25 million to Trump-aligned interests, yet GPT-5.6 remains sidelined.
This federal bottleneck is accelerating a strategic backfire. New reports indicate Chinese models are reaching parity; 360 Security Technology’s tool using GLM 5.2 matched Mythos in finding cybersecurity bugs. On The AI Daily Brief, Nathaniel Whittemore noted that Coinbase has already cut its AI bill in half by defaulting to Chinese open-weight models like GLM 5.2 and Kimmy 2.7.
“If the rest of the world builds on a Chinese tech stack because the U.S. version is walled off, American AI sovereignty becomes a hollow victory.”
- Nathaniel Whittemore, The AI Daily Brief
The capability gap is closing fast. On Moonshots, Emad Mostaque projected Chinese open-weight models could match U.S. frontier performance by December. Analysts warn the U.S. is withholding the best defensive tools from its own public, potentially leaving domestic infrastructure more vulnerable.
The long-term risk is a permanent intelligence hierarchy. Andrew Curran predicts a core structure of restricted access will endure, giving the U.S. government and selected companies first access to future models. As Theo from Nerd Snipe put it, returning to older tools after tasting a leaked frontier model feels like a significant regression. The state is building a world where it holds an N+1 intelligence advantage over markets and the public, turning the democratic promise of AI into a permissioned privilege.
