Trump's AI U-turn: New executive order requests voluntary pre-release vetting of frontier models by the NSA - goes further than Biden's approach but lacks teeth
What it really says
On June 2, 2026, President Trump signed the executive order 'Promoting Advanced Artificial Intelligence Innovation and Security.' At its core is a voluntary framework that asks AI developers to submit so-called 'covered frontier models' to the US government for review up to 30 days before release. The National Security Agency (NSA) receives a central role: it is tasked with developing and maintaining a classified benchmarking process to assess the advanced cyber capabilities of AI models and determine which models qualify as 'covered frontier models.' Within 60 days (by August 1, 2026), the relevant agencies must design the voluntary framework. Additionally, an AI cybersecurity clearinghouse will be established within 30 days, coordinating vulnerability scanning in voluntary collaboration with the AI industry and critical infrastructure operators. The executive order explicitly states that nothing in it shall be construed to authorize any mandatory licensing, preclearance, or permitting requirement for the development or release of AI models. In some respects, the order goes further than Biden's October 2023 AI executive order: Biden required notification of certain training runs and sharing safety test results with the Department of Commerce, but did not demand direct access to the models themselves. Trump now requests model access, but narrows the focus to cybersecurity and national security, whereas Biden's approach was broader, encompassing privacy, civil rights, and worker protections.
Our assessment
This executive order marks a notable reversal in the Trump administration's AI policy. In January 2025, Trump revoked Biden's AI safety order and described AI regulation as anti-innovation. That the same administration is now introducing a - albeit voluntary - review framework for frontier models shows how quickly the perception of AI risks has shifted, even in deregulation-minded circles. However, the order has structural weaknesses. First, the framework is voluntary. No company is compelled to submit its models. Critics like Representative Don Beyer note that the order 'continues to allow companies to release powerful models to the public without undergoing meaningful evaluation or review.' Second, the central role of the NSA - an intelligence agency with a well-documented history of mass surveillance - raises trust concerns. AI companies may hesitate to hand their models to an intelligence agency, and the classification of evaluation criteria makes the entire process opaque. Third, binding safety standards and independent oversight are absent. For people worried about uncontrolled AI development, the message is mixed: the US government acknowledges the problem but offers only a toothless instrument.
Relevance for Germany
For Germany and Europe, this executive order has several implications. First, the contrast between the voluntary US framework and the mandatory EU AI Act highlights the fundamental difference in regulatory philosophies: Europe relies on binding rules with fines of up to 35 million euros or 7 percent of annual turnover, while the US relies on voluntary cooperation. Second, NSA involvement raises questions for European AI companies: anyone wanting to offer frontier models in the US might need to provide model access to a US intelligence agency - a data sovereignty concern that could become relevant for German providers like Aleph Alpha or SAP AI. Third, the order demonstrates that even the most regulation-skeptical US administration in recent decades is no longer relying on pure deregulation when it comes to AI. This strengthens the position of European regulators, who are often criticized as overreaching. For citizens in Germany, the order means little in concrete terms for now, but it signals that AI safety has arrived as a priority in Washington - albeit with a different focus than in Brussels.
Fact check
The primary source is the full text of the executive order on whitehouse.gov and the accompanying fact sheet, both published June 2, 2026. The core provisions - the voluntary 30-day review framework, the NSA's central role in classified benchmarking, the explicit rejection of mandatory licensing, and the establishment of an AI cybersecurity clearinghouse - are consistently reported by PBS News, CNBC, Scientific American, Crowell & Moring, Freshfields, Ropes & Gray, and the Council on Foreign Relations. The comparison with Biden's October 2023 executive order is analyzed in particular by Scientific American and the CFR. Representative Don Beyer's criticism is cited by PBS News and multiple legal analyses. The statement that the NSA determines the threshold for 'covered frontier models' through a classified process comes directly from the executive order text.
Source
- • https://www.whitehouse.gov/presidential-actions/2026/06/promoting-advanced-artificial-intelligence-innovation-and-security/
- • https://www.whitehouse.gov/fact-sheets/2026/06/fact-sheet-president-donald-j-trump-promotes-advanced-artificial-intelligence-innovation-and-security/
- • https://www.pbs.org/newshour/nation/trump-signs-executive-order-that-allows-voluntary-federal-vetting-of-top-ai-models-for-national-security-risks
- • https://www.cnbc.com/2026/06/02/trump-executive-order-ai.html