On June 2, 2026, President Trump signed Executive Order 14409, requiring the federal government to produce a framework for reviewing frontier AI models before release. The deadline was August 1. The deadline passed. The White House says the framework is classified.
Two AI model releases have already been suspended under whatever process the government is using. There is no published threshold that triggers a suspension. There is no published timeline. There is no published appeals process.
What EO 14409 required
The executive order directed multiple agencies to build a classified benchmarking process to identify which AI systems qualify as covered frontier models, and to establish a voluntary framework giving the government up to 30 days of advance access to those models before release to other partners. The framework was supposed to be published by August 1, 2026. Publishing would have created a defined, predictable process that AI developers could plan around.
What happened
As of August 1, the federal government had published no Federal Register notices, no NIST publications, no CISA advisories, and no OSTP statements related to the framework. The White House subsequently communicated that the framework exists but will remain confidential. TechPolicy.Press published five specific questions the government has not answered: what triggers a review, how long reviews can last, what criteria determine clearance or suspension, what recourse developers have, and how the process interacts with existing export controls and NSPM-11.
The suspensions that already happened
Before the framework was supposed to be published, the government intervened in two model releases without a published process. Anthropic's Claude Fable 5 and Mythos 5 were suspended globally for roughly three weeks. OpenAI's GPT-5.6 was restricted to government-vetted partners for 12 days. In both cases, developers, researchers, and enterprise customers had no published criteria to understand why the intervention happened or when it would end.
What this means for organizations building on AI
If you are deploying applications on top of frontier AI models, you are now exposed to a new category of operational risk: suspension of your upstream model without notice or timeline. This is not a theoretical risk. It has already happened twice. Neither suspension had a published endpoint criterion, meaning organizations could not plan for or around the interruption. Standard business continuity planning for AI-dependent applications needs to account for this.
The Pacing the Frontier letter
On July 28, OpenAI and Anthropic signed a joint public letter requesting that the US government help build international tools capable of deliberately slowing automated AI development. Read alongside the suspended releases and the secret framework, the letter signals that the two largest closed-model labs now view government involvement in frontier AI releases as an acceptable, or at least inevitable, constraint. This is a structural shift in how frontier AI governance works.
Gigia Tsiklauri is a Security Architect and founder of Infosec.ge. Get in touch if you are a security or compliance team navigating AI governance risk and want to talk through what controls and contingency planning look like for AI-dependent applications.