AI Confrontation: Washington and Beijing Shift from Rivalry to Direct Clash

The period has been marked by a crystallization of Sino-American tensions in the artificial-intelligence sector, moving from the rhetorical arena to direct accusations and concrete industrial pressure. The United States now publicly accuses Chinese firms of systematically stealing AI models, while the American automotive sector is asking Congress to ban Chinese connected vehicles—a concern shared by Germany. This confrontation is intensifying just as China demonstrates spectacular capability advances, notably with a first autonomous submarine mother ship, prompting global security concerns and a regional arms race in Southeast Asia.

In parallel, the reliability of AI systems remains a point of fragility, illustrated by massive vehicle recalls, public malfunctions and growing use for malicious purposes. A report from the AI Safety Institute reveals that more than half of the models it has assessed are open-weight, and that 58% of these are of Chinese origin, underscoring the scale of the governance challenge posed by the proliferation of powerful technologies that are difficult to control.

The Sino-American Technological Rivalry Enters a Phase of Direct Confrontation

The strategic competition between the United States and China over artificial intelligence has clearly hardened over the past month. Tensions, hitherto concentrated on semiconductors, are now explicitly extending to software applications and key industrial sectors.

The strongest signal is a formal accusation from the United States, which charges Chinese companies with the systematic theft of AI models through distillation (OECD AI / GPAI, News, 08/09). This technique, which consists of using a high-performing model to train a smaller one, lies at the heart of the rapid diffusion of AI capabilities. This public accusation shifts the debate from mere competition to questions of intellectual property and industrial espionage.

The automotive sector has become the main battleground. American manufacturers have publicly urged Congress to ban Chinese connected vehicles equipped with AI (OECD AI / GPAI, News, 03/09). Their concerns are twofold: economic competition in the face of the very rapid development of the Chinese automotive industry thanks to AI (OECD AI / GPAI, 31/08), and national security. US authorities have launched an investigation into the security risks posed by Chinese-made Lidar sensors integrated into autonomous vehicles (OECD AI / GPAI, News, 26/08). This wariness is shared in Europe, where Germany is likewise assessing the espionage risks associated with Chinese AI-equipped cars (OECD AI / GPAI, News, 30/08).

Chinese Shows of Force and the Militarization of AI in Asia

Alongside this rise in tensions, China continues to showcase its technological advances, particularly in the military and robotics domains. The most significant announcement is the construction of the world's first unmanned submarine mother ship, entirely managed by an AI (OECD AI / GPAI, News, 19/09). This advance underscores the development of sophisticated autonomous capabilities of strategic reach.

These developments are fueling global and regional security concerns. The proliferation of Chinese AI-driven humanoid and quadruped robots has triggered an international policy debate on potential military threats and the associated ethical questions (OECD AI / GPAI, News, 16/08 and 25/08). In direct response to tensions in the South China Sea, the Philippines have announced their intention to expand their fleet of AI-equipped military drones (OECD AI / GPAI, News, 25/08), illustrating a dynamic of technological arms racing in the region.

Despite these advances, an adviser to the Chinese central bank issued an internal warning, fearing that AI might aggravate the country's economic imbalances (OECD AI / GPAI, News, 19/09), suggesting that the socio-economic implications of this rapid transition are being debated at the highest level in Beijing.

Model Proliferation and the Persistent Fragility of AI Systems

AI governance faces a twofold challenge: the rapid proliferation of models and the persistence of security and reliability flaws. The AI Safety Institute (AISI) has published key data on the models it has tested, revealing that more than half are open-weight, which facilitates their replication and modification. Notably, 58% of these open-weight models are of Chinese origin (디지털투데이, 17/09). This figure confirms China's predominance in the open-source ecosystem and complicates control efforts, a trend already observed by the AISI in previous months (contexte — 06/2026).

The period was punctuated by incidents recalling the fragility of these technologies: * The virtual actress Tilly Norwood, powered by an AI, experienced a spectacular malfunction during a live interview, unexpectedly beginning to speak Chinese (OECD AI / GPAI, News, 18/09). * Tesla had to recall nearly three million vehicles in China owing to defects in its AI-based driver-monitoring system (OECD AI / GPAI, News, 22/08). * A humanoid robot also malfunctioned and collapsed during a public event in Beijing (OECD AI / GPAI, News, 18/08).

These incidents, whether safety-critical or anecdotal, continue to erode confidence in the technology's maturity, a sentiment reinforced by the systemic alerts of previous months (contexte — 07/2026).

Cybersecurity and Malicious Uses: The Threat Materializes

The use of AI for malicious purposes is intensifying. Chinese hackers have reportedly doubled their volume of cyberattacks using the DeepSeek AI (OECD AI / GPAI, News, 24/08). On the disinformation and fraud front, AI-generated images have been used in France in a sophisticated scam targeting the families of deceased firefighters (OECD AI / GPAI, News, 13/08). These developments confirm the materialization of the risks anticipated by regulators and security institutes. In response, the European Union announced the establishment of a new team dedicated to combating deepfakes, illicit images and hacking (South China Morning Post, 31/07).

Faced with the escalation of risks, voices are being raised calling for de-escalation and cooperation. Experts have issued a pressing appeal for a dialogue between the United States and China on AI safety and its military use (OECD AI / GPAI, News, 15/09). This appeal was echoed by influential figures such as Bill Gates, who warned about the risks of AI while urging Sino-American cooperation (OECD AI / GPAI, News, 09/09).

To Watch

• China's response to the American accusations of AI-model theft through distillation.
• The follow-up to the US investigation into the security risks associated with Chinese Lidar and the debates in Congress over a possible ban on Chinese connected vehicles.
• The AI Safety Institute's (AISI) upcoming publications on model evaluation, notably on the cybersecurity risks of open-weight models.
• The first actions of the new European team combating deepfakes and how it will interact with platforms and model providers.