Beijing Grapples With Controlling Its Domestic AI Sector

China is increasingly concerned over what its AI investments have wrought.

On July 7, Reuters reported that Beijing is considering placing export controls on Chinese AI models, mirroring a move by the U.S. Commerce Department in early June to place controls on Anthropic’s latest products due to their hacking capabilities. Those controls have since been removed.

The discussion showcases Beijing’s growing concern over the cyber capabilities of its AI sector as it grapples between its open-source strategy, which prioritizes global diffusion and ease of access, and the dangers presented by more advanced models.

Commerce Ministry Considers Controls on Closed and Open-Source Models

China’s Commerce Ministry has reportedly held meetings with several leading Chinese AI firms, including Alibaba, ByteDance, and Z.ai to discuss the possibility of new export controls on both closed and open-source models — a restraint complicated by the inherently public and diffused nature of the latter product.

The proposed controls included both simple licensing regimes for lower-end models and more extensive control and monitoring regimes for higher-end products. According to Reuters, the ministry also focused on placing more stringent criminal penalties on attempts to foreign or domestic efforts to distill Chinese models. U.S. frontier labs have previously accused Chinese firms of conducting distillation attacks, a type of intellectual property theft, against their own products.

The proposal follows Beijing’s recent efforts to tighten control over domestic technology firms, including introducing new regulations in June to restrict outbound investment and expand export controls. The discussions also occurred as Chinese authorities continue to investigate Manus, a Chinese AI startup connected to American firm Meta, for potential violations of export control law due to its efforts to exit the country and evade Beijing’s control.

Washington and Beijing Weigh New AI Strategies

The Chinese debate over potential export controls for its AI models reflect a potential division between Beijing’s growing confidence in the strength of its domestic open-source labs and its concerns over the capabilities of their highest-end products. Responding to Chinese state-aligned cybersecurity firms calling for domestic labs to develop their own “Mythos-class” models, the system developed by Anthropic reportedly capable of rapidly identifying cyber vulnerabilities, Chinese firm Z.ai has publicly deployed products with similar capabilities. Moreover, Chinese open-source models have steadily gained popularity in the United States due to their lower cost, with Microsoft reportedly considering offering a domestically hosted version of DeepSeek as a possible alternative to its other AI products. 

The United States has also moved toward introducing export controls on its most advanced AI models, albeit with limited progress. On June 2, the Trump administration issued an executive order that called on American AI firms to submit their models to a voluntary 30-day pre-public testing period to identify their capacity to exploit possible cybersecurity vulnerabilities. This voluntary regime appeared to collapse later in the month as Anthropic publicly released Fable, a reportedly safer version of Mythos, without submitting to the testing period. In response, the Commerce Department placed extensive export controls on both models, barring all foreign nationals from accessing the products, including those working within the firm. While the Commerce Department eventually canceled all restrictions on Fable and Mythos after three weeks of reported consultations between Anthropic and U.S. officials, the Trump administration has attempted to codify this pre-screening process in its discussions with OpenAI, another frontier lab.

U.S. Should Pursue Responsible Deployment Among Trusted Allies and Partners

As with Washington’s efforts to develop trusted supply chains for AI via its Pax Silica initiative, the Commerce and Treasury departments should consider methodically employing controls over the diffusion of advanced AI models that pose a significant cybersecurity risk. This effort should focus on developing a standardized approach to sharing capabilities with allies and partners in advance rather than diminishing trust via self-induced whiplash — a trend that has only driven more American firms to use Chinese models.

Beijing’s discussion of regulating its most advanced domestic models due to their capabilities also highlights the continued importance of reducing China’s access to cutting-edge chips by tightening export controls. Both Congress and the Trump administration should work with key allies, particularly Japan and the Netherlands, to prevent China from acquiring high-end chip manufacturing equipment that would supercharge its domestic AI sector.

Jack Burnham is a senior research analyst in the China Program at the Foundation for Defense of Democracies (FDD). For more analysis from Jack and FDD, please subscribe HERE. Follow FDD on X @FDD. Follow Jack on X @JackBurnham802. FDD is a Washington, DC-based, nonpartisan research institute focusing on national security and foreign policy.