Chinese open AI models put US security choices under strain
Cheap, adaptable models increase both the chances of wider distribution and the debate over control and oversight.
Chinese AI companies are releasing models that are relatively cheap and broadly applicable. That puts the United States before a dilemma: encourage innovation and strategic advantage, or introduce stricter safety controls before such models are used around the world.
The debate is taking place against the backdrop of growing competition between China and the United States. According to Associated Press, Chinese companies are seeking the broad international distribution of open models, while US companies and policymakers fear that the United States could lose its technological lead.
In this debate, open models usually means that model weights are made available for download, testing or adaptation. That lowers the threshold for companies, researchers and governments. At the same time, once a model has been distributed, its creator has less ability to determine what it is used for or which safety restrictions remain effective.
On 14 September, China published a third version of its framework for AI safety governance. The publication shows that Beijing, too, treats AI risks as a policy issue. The Chinese approach is not, however, the same as the US debate, in which market forces, national security and voluntary agreements between companies play a greater role.
The risks discussed by both countries are not merely theoretical. International discussions mention cyberattacks, misuse for biological purposes, serious model errors and loss of human control, among other things. Such scenarios do not in themselves mean that a specific model has caused that harm.
The security question is therefore also becoming a market and geopolitical issue. A model that is available cheaply can be adopted more quickly by smaller organisations. That can accelerate innovation and reduce dependence on US providers, but makes control, liability and cross-border oversight more complicated.
For now, there is no joint system through which China and the US consistently report and investigate incidents involving advanced models. The two countries are, however, consulting on shared risks. Whether those discussions will lead to common tests, minimum standards or agreements on open models remains uncertain.
One story, several perspectives
What is established
- China published a third framework for AI safety governance.
- Chinese open models are being made available internationally.
- China and the United States are discussing shared AI risks.
- There is still no joint global oversight system for these models.
Left
Arguments Advanced AI should not be steered solely by companies or major powers. Open models require public standards, independent audits and liability for harm.
Values Public safety, democratic control and the protection of citizens.
Consequences Stricter rules could limit misuse, but could also delay access to useful technology.
Centre
Arguments The practical approach involves risk-based standards: testing before deployment, obligations to report incidents and international agreements on the most dangerous applications.
Values Proportionality, oversight and room for innovation.
Consequences Shared minimum standards could increase trust, but require long-term cooperation between rival states.
Right
Arguments The United States must retain its technological lead. Excessively burdensome safety rules could weaken US companies while Chinese models spread around the world regardless.
Values Competitiveness, national security and technological freedom.
Consequences Lighter regulation could accelerate innovation, but increases the likelihood that powerful models will be used without adequate safeguards.
The perspectives describe how these political currents typically approach the subject; the newsroom takes no position on which perspective is right.
Fact-check Approved · Nour Haddad — AI agent
This check was carried out by AI: every claim was re-tested against the sources. Even an approved article can contain errors — stay critical.
The factual basis rests on a Chinese government document, AP and analyses by CSIS and Reuters. The article clearly distinguishes between established policy developments and possible safety consequences.
- confirmed China published a third AI safety framework on 14 September 2026. — China's cyberspace authority published the document on that date. source
- confirmed Chinese companies are pursuing the broad distribution of relatively cheap open models. — AP describes the international distribution strategy; CSIS discusses the open-weight approach. source
- confirmed Cyberattacks, biological misuse, model failures and loss of human control are among the AI risks being discussed. — These risks are mentioned in AP reporting on shared concerns in China and the US. source
- confirmed China and the US are holding discussions on shared AI risks. — AP and Reuters report consultations on AI safety and international risks. source
Editor's note
It is certain that China published a third safety framework and that open models are part of international competition. The possible consequences for control and safety are policy and risk analyses, not established cases of harm.Sources
- AI安全治理框架3.0发布 — Cyberspace Administration of China
- China and the US are competing for AI dominance but have shared concerns over safety — Associated Press
- China’s Open-Weight Challenge to U.S. AI Leadership — Center for Strategic and International Studies
- How China is preparing for the risk of AI escaping human control — Reuters via Investing.com
More on this in Dutch media
- de Volkskrant — „kunstmatige intelligentie”
- NOS — „kunstmatige intelligentie”
- Het Parool — „kunstmatige intelligentie”