Trump exempts Chinese AI models

The Trump administration has drawn its new AI safety-review framework around a narrow slice of the market: top-tier closed models from U.S. developers. People familiar with the policy told POLITICO on August 4 that the White House framework would focus on advanced products from companies such as OpenAI and Anthropic while exempting open-weight models. The White House has not yet published the full framework, but the approach follows a June 2 executive order that set up a voluntary review process for “covered frontier models.” That structure matters because some of the fastest-moving open-weight activity is coming from outside the companies most likely to be reviewed. The New York Times reported on August 5 that the administration’s new guidelines appear to exempt Chinese AI models while still applying in practice to leading American firms such as OpenAI and Anthropic. Reuters, cited by U.S. News and other outlets, separately reported that Trump advisers told AI firms they would not safety-test open-weight models. (politico.com) ### Which models does the White House say it wants to review? President Donald Trump’s June 2 executive order said the government would develop a classified benchmarking process to identify “covered frontier models.” A White House fact sheet issued the same day said the framework would provide the federal government with secure early access to those models through a voluntary process and said the order did not authorize mandatory licensing, pre-clearance or permits for AI releases. (nytimes.com) POLITICO reported that only “state-of-the-art” models deemed national security risks would be covered and that the government review period would be capped at 30 days. White House spokesperson Liz Huston said the framework advances the administration’s “America First cybersecurity strategy” by strengthening national security and “cementing American AI dominance,” according to POLITICO. (whitehouse.gov) ### Why does the exemption point toward Chinese and open-weight models? Open-weight models are customizable systems whose core components are publicly released. POLITICO reported that the White House policy would not apply to those models, even as open models have gained traction across the industry, including among developers in China. The New York Times said that design appears to leave Chinese models outside the practical scope of the new review system. (politico.com) The White House has not publicly released model-by-model criteria, so the exact boundary remains unclear. Axios reported on August 3 that the administration had met its deadline to establish the voluntary framework but would not say what it contained, who had seen it or when companies would start using it. ### Where does Nvidia fit into this? Nvidia has become one of the industry’s most visible backers of open-weight AI. (politico.com) POLITICO reported that Chief Executive Jensen Huang was on Capitol Hill in late July arguing against possible restrictions on open-weight models, and Nvidia has continued promoting Nemotron open models in public announcements, including a July 15 release tied to Japanese enterprises and research institutions. (axios.com) Business Insider reported on August 5 that Nvidia is staffing a new AI safety team as it doubles down on open models. Markets Insider, which aggregated the report, said the move signaled a larger investment in secure AI as Nvidia expanded its open-weight push. ### Why is Nvidia also facing antitrust questions? TradingKey reported this week that the U.S. Department of Justice has opened or intensified an antitrust investigation into Nvidia over allegations that it uses market dominance to penalize customers that buy rival chips or rely on multiple cloud providers. (politico.com) TradingKey also said investigators were examining software lock-in tied to CUDA, Nvidia’s programming platform. Reuters has not independently confirmed those details. (l.businessinsider.com) Those competition questions sit alongside the company’s growing role in AI governance debates. Nvidia is pressing for fewer restrictions on open-weight models even as it builds internal safety capacity and remains central to the compute stack that many frontier developers depend on. That combination has made the company relevant to both safety policy and market-power scrutiny. ### Why are policy analysts calling this fragmented governance? (tradingkey.com) Tech Policy Press wrote in June that the Trump administration had already expanded federal oversight over frontier AI through executive action while also rescinding other measures, including export controls that had affected Anthropic models. In a separate analysis, Tech Policy Press argued that merger law is a poor fit for frontier AI because the sector is organized through partnerships, cross-holdings and compute dependence rather than simple ownership lines. (markets.businessinsider.com) The next concrete step is publication and implementation of the White House framework. Until the administration releases the full document, the practical reach of the review system — including which companies participate and which models qualify — will remain defined mainly by briefings to industry and reporting from outlets that have seen partial details. (politico.com) (techpolicy.press)