If you needed further evidence of how powerful AI models have become or that they've risen to the level of national security concerns, then the recent actions of the world's two largest economies should be enough to convince you. The US government restricted international access to Anthropic's Mythos and OpenAI's GPT-5.6, and now China is looking to do the same for the most advanced models from its frontier AI labs.
When Washington realized that Claude Mythos could be used by attackers to find software vulnerabilities and exploit systems faster than anything we've seen before, it took the drastic step of issuing export controls on the software, which essentially locked it down for three weeks while the US government and major US corporations had time to harden their systems.
According to a Reuters report, in June the Chinese Ministry of Commerce held meetings with AI labs, including Z.ai, Alibaba, and ByteDance, to discuss restricting international access to the most advanced AI models produced by Chinese labs. The government officials also raised the possibility of limiting international investors' ability to fund Chinese AI startups.
The Chinese move could have ripple effects across the US, given the widespread use of open-weight Chinese models by US corporations and AI enthusiasts. A report from OpenRouter found that over 30% of the tokens used on Chinese models came from the US, and that number has surged as high as 46% during some weeks in 2026.
With AI costs soaring in many organizations, IT departments and workers have turned to open-weight models since you can download them and run them locally to save costs, improve performance, and maintain privacy and security. And the Chinese open-weight models like DeepSeek, Qwen, and GLM are generally considered the most advanced.
If China moves to restrict US access to these models, it could boost other open models made in the US, such as Nvidia Nemotron, Google Gemma, OpenAI GPT-OSS, and others. Meanwhile, the rhetoric continues to escalate. On Wednesday, the Chinese Ministry of Industry and Information Technology warned that Anthropic's Claude Code "contains a security back-door vulnerability that poses a serious threat." Claude Code is not officially available to use in China, but Chinese software developers have publicly stated that many use it, according to a CNBC report. And that appeared to be backed up by the fact that Alibaba banned its workers from using Anthropic tools on Monday.
Our Deeper View #
Before we get too overreactive to the US-China rhetoric, let's keep in mind that none of the recent activity is likely to stop Chinese and American technologists from using each other's models anytime soon. For example, Anthropic's Mythos is now widely available to the public. The US government's restriction simply gave agencies and corporations time to prepare their cybersecurity posture before the model became more widely available. And the Chinese government is discussing restrictions on the most advanced models. The vast majority of open-weight models would likely remain available and wouldn't affect most users. In the same way, all previous-generation Anthropic and OpenAI models remained publicly available even as the government forced the companies to temporarily restrict broader access to their latest models. In other words, American-Chinese cooperation will quietly continue beneath the surface of the latest escalation in rhetoric.