The Trump administration's restrictions on Anthropic and OpenAI have created an opening for Chinese AI firms to close part of the capability gap.
CNBC reported that Anthropic was allowed to release its Mythos 5 model to some companies and federal agencies after a two‑week shutdown while its Fable 5 model remains blocked, and that OpenAI curtailed a wider rollout of GPT 5.6 at the government's request as Zhipu this month released GLM 5.2, an open‑weight model researchers say matches top U.S. labs on some cyber benchmarks.
Corporate buyers are already shifting from unconstrained token spending to efficiency, with startups such as Lindy moving off Anthropic to Chinese provider DeepSeek, and companies including Coinbase, Shopify and Airbnb reported to be using GLM 5.2, Kimi 2.7 or Alibaba’s Qwen 3 to lower costs.
Open‑weight models are simple to download and self‑host, which speeds adoption but removes a layer of vendor control and increases operational risk.
Security researchers and vendors warn those models can automate many stages of a cyberattack and could be months away from orchestrating entire operations.
Policy remains unsettled: the U.S. has previously used export controls on chips and gear, and although Nvidia’s H200 was cleared for the China region last year the commercial flow has been limited.
Sam Bresnick, a research fellow at Georgetown’s Center for Security and Emerging Technology, called the recent developments "a pretty good wake‑up call."
