U.S. Anthropic ban opens door for open-source AI, particularly from China

The U.S. Department of Commerce ordered Anthropic to stop providing its Mythos and Fable 5 frontier models to non-U.S. nationals, according to the company. This directive prevents access for foreign nationals both outside and inside the U.S., including Anthropic’s own employees. In response, Anthropic has suspended access to these models for all users.

Why is the U.S. restricting Anthropic’s AI models?

The Department of Commerce issued the order based on U.S. export rules. Anthropic had previously stated its Mythos model was too powerful for general public release without safeguards.

To manage this, the company ran Project Glasswing, an early-access program for key institutions in about 15 countries, including Japan and South Korea, to identify security vulnerabilities. The current government order overrides these previous arrangements.

Did You Know? DeepSeek’s V4 Pro costs $3.48 per 1 million tokens of output, while Anthropic’s Fable 5 model costs $50 for the same amount of data.

How are Chinese AI labs responding to the ban?

Chinese AI labs are positioning the controversy as a public relations victory. Knowledge Atlas, known as z.ai, saw its shares surge over 30% in Hong Kong trading on Monday following the release of its open-source GLM-5.2 model.

According to the South China Morning Post, Knowledge Atlas posted on social media that “frontier intelligence should not belong to only a few people.” The company’s shares have risen more than 800% since their January debut.

Demand for Chinese models has already surpassed U.S. models on the OpenRouter platform. Last week, the top four most-used models came from DeepSeek, MiniMax, Tencent, and Xiaomi.

Expert Insight: Samantha Carter notes that the shift toward “sovereign AI” represents a strategic pivot. By reducing reliance on a single nation’s proprietary models, governments are attempting to mitigate the risk of sudden access withdrawal, effectively treating AI infrastructure as a matter of national security.

What is the impact on “Sovereign AI” and global adoption?

The restriction may accelerate the adoption of open-source models, which users can run on private computers or cloud networks to sidestep government and developer controls.

Paul Triolo, a partner at DGA-Albright Stonebridge Group, said this is the first time a government has ordered a developer to restrict a model based on nationality. He suggests companies and governments may now reconsider which partners they choose for sovereign AI deployments.

South Korea has already launched a state-backed competition to develop Korean-language models. Sung Kim, founder of the startup Upstage, called AI a “strategic national asset” during a Tuesday press conference.

How do Chinese open-source models compare to U.S. versions?

Chinese models generally lag behind U.S. technology. DeepSeek’s V4 performs similarly to Anthropic’s Claude Opus 4.6 and OpenAI’s GPT 5.4, both of which were released in February and March 2026, respectively.

DeepSeek estimates it is three to six months behind state-of-the-art frontier models. However, these models are popular in developing countries due to their price-to-performance ratio.

Neil Shah, vice president of research at Counterpoint Research, stated the ban may vindicate China’s move toward tech self-sufficiency started in 2022. He noted that while export controls keep them off the cutting edge, China has its own silicon and software.

What happens next for global AI access?

The move against Anthropic could signal a broader trend. Non-U.S. organizations may be locked out of the best U.S. models if OpenAI or Google face similar export controls.

Analysts expect a rise in non-U.S. origin models from providers like Mistral and Cohere. According to Shah, the order will likely “push scale for Chinese open-source models” and encourage Middle Eastern economies to build indigenous software models.

Japan may turn to OpenAI to strengthen its cybersecurity defenses as an alternative to Anthropic.

Frequently Asked Questions

Why did Anthropic suspend access to its models for everyone?
The U.S. Department of Commerce ordered the company to stop providing access to non-U.S. nationals. Because export rules also apply to foreign nationals inside the U.S., including employees, Anthropic decided to suspend access for all users.

Which Chinese models are currently leading in usage?
According to OpenRouter, the top four most-used models last week were from DeepSeek, MiniMax, Tencent, and Xiaomi.

What are the performance differences between DeepSeek and U.S. models?
DeepSeek’s V4 performs at roughly the same level as Claude Opus 4.6 and GPT 5.4, though the company estimates it is three to six months behind the state-of-the-art frontier models.

Do you believe national restrictions on AI models will accelerate the development of independent, sovereign AI systems worldwide?