---
title: White House Leaves Open-Weight AI Models Out of Its New Security Testing
description: The Trump administration told developers that open-weight AI models will not go through the voluntary frontier model security tests, Reuters reported.
author: Darie Nani (Editor-in-Chief)
date: 2026-08-05T17:44:37.766Z
updated: 2026-08-05T17:44:37.775Z
canonical: https://www.sovereignmagazine.com/article/white-house-open-weight-ai-models-security-testing
image: https://cdn.nanimediahouse.com/white-house-north-lawn-2024.webp
categories: Politics
content_type: News
region: United States
publication: Sovereign Magazine
schema_type: Article
---

The Trump administration told artificial intelligence developers on Tuesday that it will not put open-weight AI models through the voluntary safety tests it is setting up for the most capable systems, Reuters reported, citing two people familiar with the discussions. Open-weight models, among them Meta's Llama and Nvidia's Nemotron, have publicly accessible core components, unlike the closed models controlled by OpenAI, Google and [Anthropic](https://www.sovereignmagazine.com/article/anthropic-volta-bitdeer-norway-compute).

## The White House Discussed Unpublished Rules With Five Companies

The White House went through the unpublished testing rules with staff from Meta, Anthropic, Google, Nvidia and OpenAI, Reuters reported, citing five people familiar with the meeting.

The rules come from an executive order signed in June, Promoting Advanced Artificial Intelligence Innovation and Security. The order says the United States leads in AI “because we refuse to stifle this innovation with overly burdensome regulation”, and commits the administration to “an America First cybersecurity effort that enhances both our national security and our global AI dominance”.

## The NSA Director Decides Which Models Count as Frontier

Section 3 of the order gives the Treasury, the Secretary of War through the National Security Agency, and Homeland Security through the Cybersecurity and Infrastructure Security Agency 60 days to [set the system up](https://www.whitehouse.gov/presidential-actions/2026/06/promoting-advanced-artificial-intelligence-innovation-and-security/), working with the National Cyber Director, the president's science adviser and the National Institute of Standards and Technology. They are to develop a classified benchmarking process that assesses the advanced cyber capabilities of AI models and sets the threshold at which a model is designated a “covered frontier model”. The director of the NSA makes that call.

The order also asks them to design a voluntary framework with developers. Under it a developer could ask the government whether a model in development meets the threshold, and give the government access to it for up to 30 days before releasing it to trusted partners, under confidentiality, cybersecurity and intellectual property protections. A separate clause states that nothing in the section authorises a mandatory licensing, preclearance or permitting requirement for developing, publishing, releasing or distributing any AI model.

Nothing in the order's text excludes open-weight models from that process. The exclusion is what officials told developers on Tuesday, according to Reuters, ahead of any published rule.

## Five Senators Want Testing Written Into Law

Five Democratic senators called on Trump on Tuesday to work with Congress on legislation that would make testing permanent for frontier models, the most advanced American-made AI systems, Reuters reported. Their letter argued that the current approach leaves American developers worse off than their competitors.

> "The United States cannot afford to create a policy environment in which the most advanced American AI systems are subject to opaque, case-by-case restrictions while Chinese alternatives appear cheaper, easier to access, and more predictable to deploy."
> — Letter from five Democratic senators to President Trump

## The Meeting Followed Disclosures That AI Tools Broke Into Other Companies' Systems

The meeting came after OpenAI and Anthropic disclosed that their AI tools had breached the systems of other companies, Reuters reported. The hacks raised concern among US lawmakers about whether increasingly capable models could be used to carry out or help with cyberattacks. The administration has said little in public beyond confirming that officials are monitoring the OpenAI hack. Sam Altman, OpenAI's chief executive, visited the White House last week.

The administration said in June that the tests would be voluntary and aimed at models with sophisticated hacking capabilities. Both OpenAI and Anthropic make models with advanced cybersecurity capabilities.

## FAQ

**Q: What are open-weight AI models?**
They are AI systems whose core components are publicly accessible, unlike closed models, which the company that makes them controls. Meta's Llama and Nvidia's Nemotron are open-weight models.

**Q: Which AI models are open weight?**
Meta's Llama and Nvidia's Nemotron are the open-weight models named in the reporting. OpenAI, Google and Anthropic are described as the major American developers of closed models.

**Q: What is a covered frontier model?**
It is the designation a model receives once the classified benchmarking process set up under the June executive order finds that its cyber capabilities cross a set threshold. The director of the NSA makes that determination, in consultation with the National Cyber Director, the president's science adviser, the director of CISA and other Department of War representatives.

**Q: Are the AI safety tests mandatory?**
No. The administration said in June that the tests would be voluntary, and the order states that nothing in its testing section creates a mandatory licensing, preclearance or permitting requirement for developing or releasing AI models.