Trump AI safety plan favors incumbents

The Trump administration has drawn a line through the AI market that runs less between U.S. and Chinese companies than between closed and open models. At a White House meeting on August 4, officials told leading AI developers that a new voluntary safety-review framework would focus on top-tier proprietary systems, including the kinds built by OpenAI and Anthropic, while exempting open-weight models, according to reports from Politico, CNBC and the New York Times. The policy stems from President Donald Trump’s June 2 executive order directing federal officials to create a process for identifying “covered frontier models” and, for participating companies, giving the government access to those models for as long as 30 days before broader release. CNBC reported that the Treasury Department, National Security Agency and Cybersecurity and Infrastructure Security Agency were tasked with creating a classified benchmarking process for advanced cyber capabilities. (politico.com) The White House has not publicly released the completed framework or the thresholds that determine which models qualify for review. Politico reported that only “state-of-the-art” models deemed national-security risks would be covered, and that the benchmark and threshold are expected to remain classified. ### Which companies appear most exposed under the plan? (cnbc.com) OpenAI and Anthropic appear most exposed because the framework is aimed at advanced closed models with sophisticated cybersecurity or hacking capabilities. Reuters, via U.S. News, reported on August 4 that the tests were designed for models with advanced hacking capabilities and that OpenAI and Anthropic both have such models. Politico separately reported that the policy would focus on top-tier products from U.S. developers such as OpenAI and Anthropic. (politico.com) Google was also part of the White House discussions. CNBC reported on August 3 that OpenAI, Anthropic and Google were expected to attend the White House meeting on the framework. ### Why do open-weight models matter so much here? Open-weight models matter because the administration appears to have exempted them entirely from the new review process. Politico reported that the vetting policy would not apply to so-called open-source or open-weight models, a category that has drawn support across the tech industry and includes rising activity from Chinese AI developers. (usnews.com) The New York Times reported that the plan appeared to exempt Chinese models even as OpenAI’s and Anthropic’s did not. (cnbc.com) Chinese competition has been part of the administration’s stated rationale. Politico said officials were trying to balance security concerns against the risk that the United States could lose ground to Beijing if regulations became too heavy. CNBC also reported in July that cheaper open-weight models from China were closing the gap with American frontier labs. (politico.com) ### What does “voluntary” mean in practice? White House officials have described the program as voluntary. CNBC reported that the framework cannot be used to create a mandatory licensing or preclearance system, and a White House official said decisions on timing and scope of releases “rest entirely with the companies.” Liz Huston, a White House spokesperson, said the framework advances the administration’s “America First cybersecurity strategy” and that companies that choose to participate are “putting American innovation, security, and cyber defense first,” according to Politico. (politico.com) ### Why are traders and investors paying attention? Market participants are watching because the framework changes who may face delays, disclosure burdens or extra coordination before releasing top-end models. (cnbc.com) The New York Times described the winners of the plan as firms positioned around models that avoid the review burden, including some Chinese and open-weight systems, while larger U.S. distributors could benefit if compliance becomes easier for them to absorb than for smaller rivals. (politico.com) The next step is still opaque. As of August 6, the White House had not published the framework, CNBC said, and the classified benchmark overseen by the NSA, Treasury and CISA is expected to determine which future models are pulled into review. (cnbc.com) (nytimes.com)