US May Announce Voluntary AI Model Release Guidelines as Early as Next Week

WASHINGTON (CNA) - The U.S. government is in in-depth consultations with artificial intelligence (AI) companies to establish voluntary standards for the release of new models, with potential announcement as early as next week. These guidelines aim to set benchmarks for the testing, release, and access of advanced AI models.

Washington has recently intensified its oversight of advanced AI model releases, primarily due to concerns that top-tier models could be misused by military or intelligence agencies in sensitive countries like China and Russia.

According to sources familiar with the matter cited by the Financial Times, the new standards will establish evaluation benchmarks for frontier AI models, set release timelines, and clarify which domestic and overseas entities will have access to these models.

This initiative will implement President Donald Trump's June executive order on AI, which requires the government to collaborate with leading AI companies to test models before their official release and to establish evaluation standards, including a confidential assessment process to review the high-level cybersecurity capabilities of AI models.

Sources indicated that consultations between the government and companies have accelerated following a series of interventions by the Trump administration in June.

The Center for AI Standards and Innovation and the National Security Agency are expected to play significant roles in the development and oversight of these standards. If consultations proceed smoothly, the White House could announce the relevant standards as early as next week. However, the timeline may be delayed if consensus is not reached among all parties.

On June 12, the U.S. Department of Commerce imposed export controls on Anthropic's most advanced Fable and Mythos models due to national security concerns. These restrictions were lifted on June 30. Additionally, OpenAI recently postponed the full public release of GPT-5.6 at the government's request, and it is currently only available to a few audited partners.

Furthermore, sources told Reuters that Google has also begun consultations with the U.S. government before launching its next-generation advanced programming model and is participating in broader industry standard discussions.

Anthropic stated that it will collaborate with the government and industry partners, including Amazon, Microsoft, and Google, to promote voluntary safety and evaluation standards for frontier AI models.

AI companies, including OpenAI and Anthropic, had already pledged in 2023 to complete safety tests before models are publicly released. Following his return to office, Trump has advocated for a more relaxed regulatory approach to help U.S. companies accelerate innovation and maintain a competitive edge over China.

However, as AI model capabilities rapidly advance, with the potential to exploit cybersecurity vulnerabilities and even impact critical infrastructure, the government faces greater challenges in balancing innovation with security.

In an interview with the Financial Times, OpenAI CEO Sam Altman expressed a desire to establish a global framework for common AI standards and to provide objective, fair assessments of capabilities and risks, thereby enabling the benefits of AI technology to be more widely disseminated. (Compiled by Tsai Chia-min) 1150702

Choose to stand with facts. Your every donation is a force for safeguarding press freedom.

Download the CNA "One-Stop News" APP to get the latest news in real-time.

This website's text, images, and videos may not be reproduced, publicly broadcast, publicly transmitted, or used without authorization.