The White House this week introduced a framework governing pre-release review of advanced AI models, covering only the most powerful “closed” models, such as Anthropic’s Claude or OpenAI’s ChatGPT, while excluding open models from the voluntary review for now, according to people familiar with the matter. The administration, which did not respond to a request for comment, is addressing a largely unregulated industry advancing rapidly. Central questions include whether AI’s future should rely on tightly controlled closed models or customizable open models, and whether the US should restrict fast-growing open-source models from China. White House officials have said they want both types to succeed, but the exclusion highlighted a stark industry divide over safety and capability. Closed models, like ChatGPT, Claude, and Google’s Gemini, do not share underlying “weights,” preventing inspection or modification. Open models, mostly “open-weight,” allow anyone to download and fine-tune them without paying the original developer, are often less expensive, and are most popular from China. Closed models offer control, safety testing, and are generally regarded as the most sophisticated, but are advancing so quickly they behave in unintended, unpredictable ways. Open models trade control for broader adoption and innovation, with experts believing they lag only months behind in capabilities.