On August 5th, local time, Bloomberg reported that the White House had informed top US artificial intelligence (AI) companies that under the new AI security framework established by the Trump administration, open-source models developed by Chinese competitors would not need to undergo security tests by the US government.
According to reports, this yet-to-be-released AI security framework was developed in response to an executive order signed by President Trump in June, aimed at addressing AI-related security issues. The order included a voluntary program requiring AI companies to submit their advanced models for review by the U.S. government.
The source revealed that the decision to grant this “exemption” was communicated by U.S. officials to the industry representatives during a closed meeting at the White House on Tuesday. Companies from Silicon Valley such as OpenAI, Anthropic, and Google, which is part of Alphabet, sent representatives to attend the meeting. Neither the White House nor the representatives from these three companies commented on the matter.
Reports suggest that as Chinese developers like Moon Darkness and DeepSeek continuously release open-source systems that can rival the top-tier American ones, controversy over how to regulate open-weight models in Washington is growing.
Although American companies like OpenAI are also involved in open source models, their core business remains focused on selling closed-source or proprietary systems to customers. In contrast, Chinese companies have already launched a number of cost-effective open models into the market. Although these models are slightly inferior in terms of absolute performance compared to leading American competitors, they remain highly competitive due to their price advantages.
According to US media, excluding open weight models from voluntary security testing programs was a setback for Anthropic CEO Dario Amodei. He had called for mandatory government security reviews for both open and proprietary models, but such restrictions faced opposition from almost the entire American tech industry, including NVIDIA. Amodei also falsely accused Chinese AI models of violating US regulations, which led to responses from Treasury Secretary Sebastian and others.
In response to industry complaints about the current review methods for AI models, Bessen proposes establishing an independent AI regulatory agency, modeled after the Financial Industry Regulatory Authority (FINRA) in the United States. This agency would give companies significant input into security reviews. It is currently unclear whether this plan will be incorporated into the AI Cybersecurity Order, or whether it will be pursued as a separate bill.
According to American media reports, this meeting was hosted by the Office of the National Cyber Director of the United States. Sean Kanneh Ross, the National Cyber Director of the United States, revealed at a cybersecurity conference in Las Vegas that the government intends to support the open-source AI ecosystem in the United States, stating that such models have 'a significant role'.
He emphasized that the regulatory framework must maintain 'flexibility' to encourage information sharing between the industry and the government. 'Regulatory systems not only stifle growth, development, and innovation, but also cause significant harm to the industry. Moreover, they become outdated within 48 hours after going through any procedures.'
This stance is similar to the joint letter issued by dozens of American companies recently, which supported the open-weight AI models. 25 companies and institutions, including Microsoft, NVIDIA, Dell, IBM, Meta, Palantir, and Hugging Face, explicitly supported open-weight AI models in the letter, calling on policymakers to avoid premature restrictions or bans.
Almost all major AI laboratories and infrastructure companies in the United States have joined the petition, but Google and Musk’s xAI did not participate. Anthropic, on the other hand, is the most conspicuous absentee from the list.
