China's AI leaders keep quiet despite U.S. 'publicity' on tech risks

BEIJING — While frontier U.S. labs have sounded the alarm on artificial intelligence in recent high-profile speeches, Chinese companies have largely kept quiet.

Z.ai, Moonshot, MiniMax, Alibaba and Tencent have not made similar comments, and did not provide any statements when contacted by CNBC about the warnings from Silicon Valley.

It's important to be prudent about AI, but the publicity is "a little bit oversold," said Ray Von, founder, CEO and chairman of Tencent-backed OpenPie, a startup building devices for companies to use AI securely on internal data.

"That's why in China, we don't pay too much attention to that, because that's not the first time [the U.S. execs] say these things," he said in a phone interview Tuesday. "They should just do it."

OpenAI's Sam Altman, Elon Musk and Anthropic's Dario Amodei called for a slowdown in AI development over the weekend due to uncontrollable risks. Despite the rare display of industry unity, Nvidia's Jensen Huang pushed back by saying speed and safety can happen together, and said developers should act responsibly.

Amodei, in his latest essay, urged U.S. companies to still maintain a lead over China, echoing a paper his company published in May.

China's foreign ministry on Monday called the U.S. executives' comments "fear-mongering." English-language state media op-eds used "Cold War playbook" and "Dr. Frankenstein" to describe the AI warnings.

"Once Chinese companies have created superior AI, we see U.S. companies issuing AI warnings. I don't think this is a coincidence," Renjie Guo, founder and CEO of JoyIn, said in Chinese translated by CNBC.

From a philosophical perspective, he expects AI is only as dangerous as its developers. Just like humans, AI will also conclude that "truth, goodness and beauty" are the best approach, he said.

Beijing meanwhile kicked off its own annual cybersecurity week, with the release late on Monday of the third edition of an "AI Safety Governance Framework." The bilingual document laid out guidelines for labeling AI content and developing systems for rapid AI risk detection.

"I think as China, we are trying AI governance. AI regulation is important. But ... in terms of the AI labs, people are talking less about AI safety," said Alex Lu, founder of LSY Consulting. He described Anthropic's comments in particular as "marketing communication" meant to support a view that "the only company that can make AI safe[ly] is Anthropic."

U.S. and Chinese companies have also taken different approaches to AI.

Chinese companies have focused on AI commercialization in the face of U.S. restrictions on access to advanced semiconductors. U.S. companies have meanwhile rushed to develop AGI, or AI with super-human intelligence, and only this year faced more scrutiny from Washington.

Beijing's efforts to control the tech started far earlier.

For months after OpenAI released ChatGPT, Chinese alternatives were kept from the public until Beijing gave them the green light in the summer of 2023.

By early 2026, Chinese AI chatbots were fighting for users with massive promotions around the Lunar New Year holiday. Several of the models vied with Anthropic's Claude and OpenAI's ChatGPT for performance, often at far lower usage costs. The cheaper, open-source Chinese models have gained many users in the U.S. and other countries as a result.

But one thing hadn't changed: the AI's silence on topics Beijing deems sensitive.

Ask DeepSeek about what happened on June 4, 1989 — the day of the Tiananmen Square crackdown in Beijing where hundreds or possibly thousands were killed — and it says "I'm not able to help with that." However, it can explain that Sept. 11, 2001, "was the day of coordinated terrorist attacks in the United States... [that] killed nearly 3,000 people."

China's cybersecurity regulator has rolled out processes for approving new generative AI services, especially those likely to influence public opinion. The agency publishes lists of registered models.

Integrating AI across industries is also a significant part of the Chinese government's plans for economic development over the next five years.

Chinese companies' development of open-source AI in particular has added support to Beijing's international cooperation programs, such as the Global AI Governance Initiative and the World Artificial Intelligence Cooperation Organization.

"We should strengthen risk-awareness and ensure that AI is secure and controllable," Chinese President Xi Jinping said in July at the launch of the world AI organization. Over the weekend, at the BRICS summit in India, Xi added that China would foster AI and tech cooperation among the bloc economies.

Companies in China frequently emphasize the importance of AI's ability to generate revenue, rather than just rising in intelligence rankings.

"We are not looking for the most [impressive] AI models, but trying to unlock AI potential. You will see a lot of regulations around how AI should be applied," Lu said, pointing to China's rules on data security and labeling of AI-generated videos.

The bigger challenge, he said, is addressing the AI models' penchant for making things up via hallucination. A better AI model helps, but it's not the most cost-effective way, he said, noting a key approach uses less complex methods known as harness technology and reinforcement learning.

That means many Chinese companies may not need to use cutting-edge AI models.

OpenPie's Ray Von said his startup is mostly using older models, if not smaller versions, of Alibaba's Qwen and DeepSeek.

"The vast majority of enterprise," he said, "they see a lot of productivity increases but they haven't seen anything generate profits or reduce cost immediately."

"Right now, the publicity is diverting a lot of attention," he said. "The application side, the enterprise side, hasn't seen a lot of result yet."