US government creates a classified framework for security reviews of frontier AI models
According to Tech Policy Press, the US government is developing a classified framework under Executive Order 14409 for reviewing frontier AI models before release, without published criteria. The process led to a 19-day block on models from Anthropic and a limited launch of GPT-5.6 from OpenAI.
In June 2026, President Donald Trump signed Executive Order 14409, titled “Promoting Artificial Intelligence Innovation and Security". Among other things, it directs a group led by National Security Agency to create a framework under which AI developers would have to provide the government with access to their models for up to thirty days before release. According to the White House, there will be no obligation to publish the evaluation criteria, and the framework as a whole will remain secret.
According to the article, the framework was due to be completed during the week the text was written (early August 2026), but the author says the process had been operating informally for the preceding eight weeks — without published criteria, set deadlines or a clear legal basis, relying only on government pressure and calculations by companies that resistance would cost them more than cooperation. During this period, frontier models from Anthropic were put on hold for nineteen days through an export control order, GPT-5.6 from OpenAI had a two-week limited launch, and a limited group of organizations was selected for access without the selection criteria being published.
The author, an AI governance consultant, proposes a set of questions against which the resulting framework should be assessed — for example, what exactly makes a model “covered" by review (which capabilities trigger review) and how long a review may last, including what happens once the deadline expires. She cites export controls as a precedent, where thresholds (e.g. encryption bit length) are public while evaluation methods remain classified. She also mentions legislative proposals: Great American Artificial Intelligence Act 2026 (GAAIA), which would introduce transparency, independent audits and safety testing, and AI Incident Reporting Act, which would require the Department of Commerce to inform Congress after receiving a report of a serious incident.
The full source text is not available. Details can be found in the source article.
Why it matters
For companies developing frontier AI models, the absence of published criteria and deadlines means unpredictability when planning product releases — as the nineteen-day block on models from Anthropic and the two-week limited launch of GPT-5.6 from OpenAI show, delays may occur without a clear reason or end date. For users and customers, this means that access to new frontier models may be delayed or restricted to a selected group of organizations without prior notice.
Two audiences, two different impacts
What this means
For individuals
Access to newly released frontier models may be delayed or restricted to selected organizations without publicly known reasons, as happened with GPT-5.6.
More practical updates →For a business
Companies developing frontier AI models face informal government review without published criteria or deadlines, which may lead to releases being blocked for several weeks (19 days for models from Anthropic) and increases risks for product planning and compliance.
Risks and complianceCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.