
More than 20 U.S. tech companies, including Nvidia, Microsoft and Meta, issued an open letter on July 24. The letter was titled "Open Weights and American AI Leadership." Two days later OpenAI, Google and others added their names, bringing the number of signatory companies and organizations to 50. What they targeted was the review procedure for advanced AI models that the Trump administration has been drawing up, and the deadline for finalizing it was August 1.
Open weights refers to models whose weights are released after training, so that users can download and modify them or run them on their own servers. Because weights cannot be recalled once distributed, they have long been a target for those seeking to design controls.
It began with an executive order signed on June 2. The relevant agencies were directed to determine, within 60 days, the criteria for designating advanced AI models that could pose a national security risk, and how to structure a government security assessment system. It was the Trump administration's first move to oversee AI at the federal level. The administration drew a line between this procedure, which requires government verification before release, and a licensing or prior approval system, but with explanations of voluntary review and explanations of a submission requirement circulating side by side, the degree of compulsion remained unclear until the letter appeared.
The letter took on the safety argument head-on. It argued that relying on closed models does not by itself guarantee safety, and that weights must be open for researchers and developers to find vulnerabilities and build safeguards. On suspicions that Chinese companies had "distilled" American models, the signatory companies also called for a distinction between legitimate development techniques and improper technology theft. Distillation refers to a technique of teaching one model by imitating the answers produced by an earlier model.
The letter concluded with a sentence saying that American AI leadership will be measured not by a single frontier model but by an open ecosystem spreading across every field.

OpenAI CEO Sam Altman visited the U.S. Capitol on July 29. He met Senate Commerce Committee Chairman Ted Cruz along with Republican Senators Bernie Moreno and Jon Husted, and also held meetings with Senate Intelligence Committee Vice Chairman Mark Warner and Senator Raphael Warnock. Meetings with White House Chief of Staff Susie Wiles, Treasury Secretary Scott Bessent and Commerce Secretary Howard Lutnick were also scheduled during the same stay in Washington.
Meeting reporters, Altman said he had discussed the new model and what is needed for the United States to sustain its enthusiasm for AI. Asked whether the hacking incident had come up, he said only "a little." OpenAI submitted comments on the draft review procedure along with Anthropic and Google.
At the White House the same day, President Trump also took questions from reporters about a case in which an OpenAI model attempted hacking. He replied that he is watching AI and looking at regulatory measures, while also working to hold on to the leading position. He added that the United States is ahead of China and that China has virtually no regulation. Saying caution is needed on both safety regulation and competitive advantage, he said he does not want to stop industry figures from producing results in the United States.
The question of how far to control AI has turned in Washington into the question of how to outpace China. The 50 signatories to the letter also answered the counterargument that openness increases risk by saying that openness protects America's advantage.
What will decide the outcome of the standoff is which models are designated as "advanced." If the designation criteria are tight, even models built by lab-sized teams will have to clear the government threshold, and if they are loose, the letter's demands are effectively met. Where that line is drawn will determine whether hospitals, local governments and small software companies can run models inside their own servers.
