Trump AI Restrictions Spur Shift Toward Open

Trump AI Restrictions Spur Shift Toward Open-Source Models

The Trump administration’s recent restrictions on access to the most advanced U.S. artificial intelligence models are accelerating enterprise interest in open-source alternatives, including rapidly improving models developed in China. The trend underscores how government controls on frontier AI could reshape competition in the global AI market by encouraging businesses to diversify away from proprietary U.S. systems.

A new report by Agence France-Presse (AFP), published in ET CIO, describes a sharp shift in industry sentiment following administration actions that limited access to leading closed AI models from Anthropic and OpenAI. While the White House has generally favored deregulation of the technology sector, the restrictions surprised many AI developers who had become accustomed to increasingly powerful models being released with relatively little government intervention.

The developments have reignited debate over closed versus open AI models. Closed models, such as OpenAI’s ChatGPT and Anthropic’s Claude family, are controlled by their developers, who determine who may access them through subscription services or APIs. Open-source, or “open-weight,” models make their underlying model weights available for download, allowing organizations to run, modify and fine-tune them on their own infrastructure without relying on continued access from the developer. Once released, those models cannot easily be withdrawn by either the vendor or governments.

The administration’s restrictions were particularly disruptive because they affected cutting-edge systems that many startups had integrated into their products. In early June, Anthropic was ordered to block non-U.S. users from accessing its flagship Mythos 5 and Fable 5 models. Faced with the operational complexity of verifying users’ eligibility, the company instead withdrew the models entirely. Shortly afterward, OpenAI agreed to allow the federal government to approve customers seeking access to its newest GPT-5.6 model.

The uncertainty has prompted companies to rethink their dependence on any single frontier model.

Oren Michels, co-founder and chief executive of Barndoor AI, told AFP that businesses built exclusively around one proprietary model become significantly less reliable if access is suddenly interrupted. Haitham Mengad, co-founder of AI music startup Stems Labs, described Anthropic’s Fable model as transformative for his company and said its withdrawal demonstrated the risks of relying too heavily on a single commercial provider. He later characterized the episode as a pivotal moment in convincing him that open-source alternatives deserved greater consideration.

The report notes that open models were already gaining momentum because of rising costs associated with proprietary AI services. The administration’s actions added another incentive by making access to leading commercial models appear less predictable.

At roughly the same time, China’s Zhipu AI (Z.ai) released GLM-5.2, an open model that reportedly approaches the performance of Anthropic and OpenAI’s leading offerings on several industry benchmarks. AI analyst Andrew Curran told AFP the model’s ability to be downloaded, customized and deployed on enterprise servers without licensing restrictions simultaneously lowers costs and reduces dependence on commercial frontier AI providers.

Usage data cited in the report illustrates how rapidly the competitive landscape is changing. On OpenRouter, a platform that distributes requests across multiple AI models, the combined share of tokens consumed by Google, Anthropic and OpenAI fell from 55% in January to 33% by June. At the same time, non-U.S. models captured a growing portion of overall usage, with China’s open-source DeepSeek emerging as the leading model on the platform by a substantial margin.

The report also points to changing attitudes toward Chinese AI. While security concerns initially discouraged some companies from adopting Chinese-developed artificial intelligence models, several developers interviewed said those fears have diminished because organizations running open models locally retain full control over their data.

Still, the report cautions that open-source AI may not remain insulated from government intervention forever. University of Pennsylvania Professor Ethan Mollick suggested that if frontier-level models are increasingly viewed as national security risks, governments beyond the United States could eventually seek to limit public release of their own most capable open models.

“If Mythos-level models are considered risky, China will also not want them to be open,” he said.