Keith Garrett

The Trump administration is preparing an executive order, expected this week, that would create a framework for AI labs to share new models with the US government up to 90 days before public release. The NSA would handle classified testing of these models before deployment. The move was triggered by the release of Anthropic's Mythos Preview and OpenAI's GPT-5.5, both of which demonstrated advanced capabilities in identifying vulnerabilities across computer networks, including critical infrastructure systems. According to Axios and Nextgov, the national security apparatus was spooked by what these models can do and pushed for pre-release government access. The framework is described as "voluntary," but there is internal disagreement within the administration over how far to go. Intelligence agencies and the Commerce Department are fighting over who should handle model evaluation. Some officials want stronger safeguards while others want a hands-off approach to avoid slowing innovation. The key question is what "voluntary" means in practice when the entity asking for access is the NSA. Giving intelligence agencies a 90-day preview window into every frontier AI model before the public sees it is a significant expansion of government oversight over private technology development, regardless of what label gets put on it.

One week after opening pre-orders, these are still selling extremely fast. We're now into Batch 10, which is our last August batch. We're going to do everything we can to scale manufacturing capacity to build these faster too!