As the world debates the risks of AI, China closes the technology gap with the US

HONG KONG (AP) — As the rapid development of artificial intelligence triggers a renewed debate about its risks, the United States and China keep racing to develop increasingly powerful models even as they express their own concerns.

The U.S. has outpaced China and other countries for years in AI model capabilities, research, citations and other areas. But China has been rapidly catching up as an AI superpower even as U.S.-led restrictions have blocked China from accessing some of the most advanced technologies, including the cutting-edge chips and the machines to make them.

Experts and recent reports, like one by the Stanford Institute for Human-Centered Artificial Intelligence, said China has emerged as an AI counterweight to the U.S. and this year appears to be erasing the U.S. lead.

“The gap between U.S. and Chinese models is narrow and fragile,” said Samm Sacks, a senior fellow at the Johns Hopkins School of Advanced International Studies’ Institute for America, China, and the Future of Global Affairs.

The current debate takes on greater importance ahead of Chinese President Xi Jinping’s expected visit to Washington this month to meet U.S. President Donald Trump, and as experts say the future of AI will depend significantly on what China and the U.S. do and don’t do.

Here’s a quick look at some of the most important Chinese AI models and the concerns in both Washington and Beijing:

Moonshot AI leads strong competition from Chinese companies

Beijing-based startup Moonshot AI made a splash with the release of Kimi K3, taking Silicon Valley by surprise with capabilities close to U.S. rivals. It was ranked No. 3 globally in model intelligence in July, shortly after its release, behind Anthropic’s Claude Fable 5 and OpenAI’s GPT-5.6 Sol, according to AI benchmarking platform Artificial Analysis. But Kimi K3 now ranks ninth, behind new frontier models from Anthropic, OpenAI, Meta and SpaceXAI, indicating China’s AI lead is fluid.

AI startup darling DeepSeek was the first to upend global stock markets by proving Chinese AI models are capable of challenging the U.S. lead. The company’s R1 model, released in January 2025, was more cost-effective than similar U.S. models. In April, DeepSeek previewed its powerful V4 model, saying it has big improvements in knowledge, reasoning and “agentic” capabilities, referring to the ability to plan and execute multiple, complex steps with limited human guidance.

The powerful GLM-5.2 from startup Z.ai that rolled out in June raised Silicon Valley eyebrows with comparisons to the most advanced Anthropic and OpenAI models. Z.ai then launched GLM-5.3 in August with improved coding and agentic capabilities that pushed it up the global AI model rankings.

In July, China’s Alibaba previewed its Qwen3.8-Max model, which the tech giant said was its most powerful model in the series, challenging Anthropic’s Claude Fable and OpenAI’s GPT.

China worries about AI risks to security and governance

China’s state security minister has warned artificial intelligence poses a range of potential threats to the country, from political and ideological security to cyberattacks, data leaks and challenges to governance.

In an article published Sunday, Chen Yixin said the technology could threaten China’s political, institutional, and ideological security if used by “hostile forces and individuals with ulterior motives.”

Some individuals might use AI-generated deepfake audio and video to create “political rumors” on a massive scale, he wrote.

Chen specifically mentioned the capabilities of Anthropic’s Claude and OpenAI’s GPT as examples of how AI could be used to more easily identify and “weaponize” cybersecurity vulnerabilities. He warned certain nations and groups can now “execute complex hacking tasks.”

Top Chinese leaders have said China should accelerate the AI development and pursue technological self-reliance. But they also noted the need to control the risks and establish a global framework for AI governance.

US accuses China of extracting American model capabilities

The U.S. has accused China’s AI developers of engaging in what it has called “industrial-scale distillation,” in which AI models are trained to copy or imitate the capabilities of more powerful ones.

In the most comprehensive report by the government to date, the FBI and the National Security Agency and Cybersecurity and Infrastructure Security Agency said Sept. 8 that Chinese AI developers extracted capabilities from some of the most advanced AI models in the U.S. in “aggressive, malicious, and targeted distillation activities.”

Also on Sept. 8, U.S. Treasury Secretary Scott Bessent told an event that “the Chinese distill our models and they can never get ahead of us.”

China has rejected the claims and said distillation is commonly used across many AI companies. It asked the United States to refrain from making “unfounded accusations” and accused the U.S. of pursuing a “monopoly of the AI industry.”

Distillation is a standard technique used across the AI industry, but the U.S. agencies leveled a more specific accusation that Chinese firms used unauthorized access and other methods to extract restricted capabilities from American frontier models, said Lizzi C. Lee, a fellow at the Asia Society Policy Institute’s Center for China Analysis.

“These are still contested allegations at this point but I think the strategic implication is already important,” she said.

The United States spent a lot energy into controlling the inputs into frontier AI, especially chips. But now “it seems that it is finally discovering that it may also have to think about the outputs of frontier AI as a form of technology transfer,” Lee said.

“The U.S. labs spend enormous amounts of compute pushing out the frontier. But Chinese developers can learn from those systems,” she said. “And Chinese companies then (can) release very capable open-weight models that spread globally.”

___

Castillo reported from Beijing.

Chan Ho-him And E. Eduardo Castillo, The Associated Press