US Considering Issuing AI Security Order Without Mandatory Model Approval Thresholds

The Trump administration is preparing to sign an executive order on artificial intelligence security, which will incorporate AI companies into the federal regulatory system through a cybersecurity information sharing mechanism, but according to informed sources, the draft has explicitly removed mandatory cutting-edge model approval clauses.

According to an exclusive report by Bloomberg on May 8th, the aforementioned executive order will authorize agencies in the United States to collaborate with AI companies, focusing on identifying and patching AI-driven cybersecurity vulnerabilities in federal, state, and local critical infrastructures.

The timing of the rollout of this executive order is uncertain, and the content may still be adjusted.

Drastic Policy Reversal

In January 2025, at the beginning of his tenure, Trump abolished the measures by the Biden administration that required AI companies to report the safety assessment results of high-risk models and made the deregulation of AI a core policy objective.

The direct catalyst for this policy shift was the discovery that the latest model by Anthropic could accurately identify network system vulnerabilities, raising a high level of vigilance among national security officials.

According to a report by The New York Times, the White House is discussing the establishment of an interagency AI working group composed of technology company executives and government officials. According to a 16-page draft obtained by Politico, the Center for AI Standards and Innovation (CAISI) under the US Department of Commerce will take the lead in pre-release model assessments, with the National Security Agency and the Office of the Director of National Intelligence also potentially participating.

Why Was the Mandatory Approval Clause Removed

The White House issued a statement to distance itself from "more stringent regulation," emphasizing that it will "continue to empower innovators in a way that prioritizes the United States, rather than bureaucracy." The final draft clearly states: the government may conduct assessments before the model is officially released, but it does not have the power to prevent the model from being marketed and does not establish any new regulatory licensing requirements for AI models. This statement is seen as a significant concession to Silicon Valley.

Before the official executive order is issued, Google, Microsoft, and Musk's xAI have signed voluntary cooperation agreements with the government, allowing the government to conduct security assessments before the model is released. The relationship between Anthropic and the government is more complex - the company filed a lawsuit last month over federal procurement restrictions, but relations have recently eased.

The debate around AI regulation reflects deeper tensions between Silicon Valley and Washington, D.C. Supporters argue that the government should have enough information before the release of cutting-edge models to assess risks; critics warn that any approval mechanism could become bureaucratic barriers to innovation and put the United States at a disadvantage in the AI race with China. The wording of the final executive order will directly affect the compliance costs and release pace of AI developers in the US market.