China updates AI safety framework, cites deception and autonomy risks

Trump and Xi set to meet in Washington as AI rivalry deepens

China’s Cyberspace Administration published the 2026 update to its “AI Safety Governance Framework” on Monday, warning that AI is “increasingly capable of performing tasks on behalf of humans, interacting with the physical world and engaging in deception.” The 2026 update is the latest version of an annually revised document that has tracked the technology’s trajectory year by year.

The regulatory effort extends beyond the annual framework. The Wall Street Journal reported that Beijing has had AI rules for years to regulate AI development and services, including requirements that chatbots echo the Communist Party’s stances and avoid disrupting social stability. The 2025 version of the framework had warned that AI “may undergo sudden, unexpected leaps in intelligence, enabling it to autonomously acquire external resources, replicate itself and develop self-awareness.” That language, the regulator wrote, could “drive AI to seek external power and pose risks of competing with humanity for control.”

The framework update coincided with remarks by a top Chinese official on Sunday, who described AI as the “core engine of a new round of technological revolution” and the main battleground of the country’s rivalry with the United States, the Journal reported.

Chinese researchers at major AI labs told the Journal they continue to trail American labs by months or more, even as several Chinese AI models climb global rankings. That gap, some said, makes warnings of existential risk feel remote to engineers focused on closing it. The Journal reported that those building and regulating China’s leading AI models “don’t think they have built the 21st century’s version of a nuclear weapon yet—and they aren’t going to slow down while America refines its own.”

“AI progress is happening faster and it’s more frontier in the U.S., so people at U.S. labs may see more dangerous things than people in Chinese labs,” Zilan Qian, a researcher at the Oxford China Policy Lab, told the Journal. Qian added that Chinese researchers lack the same culture of evaluating societal impact that prevails in American labs.

Chinese leader Xi Jinping addressed both themes at a national AI conference in July. “We should ensure that AI is always under human control,” Xi said, calling on countries to “jointly oppose overstretching the national security concept in the field of AI and placing one country’s security over that of others.” The remarks were read as a veiled jab at the U.S. policy of banning the export of cutting-edge AI chips to China—one reason U.S. labs have moved ahead of Chinese ones, the Journal reported. Xi has also said AI will determine whether China can compete globally and has pushed the country to build a self-sufficient AI ecosystem.

President Trump has made his determination to keep the U.S. ahead equally visible. “There is a SICK conspiracy going on against AI and Data Centers, and the only one that is happy about it is China,” Trump posted on social media Monday. “We are leading China, and all others, and will continue to do so.” Trump and Xi are set to meet in Washington later this month, with officials on both sides working on the AI portion of the agenda.

Beijing has emphasized a different set of risks. State security minister Chen Yixin wrote in state media Sunday that AI poses threats—especially if exploited by foreign adversaries—including threats to Communist Party rule, critical infrastructure, data security and the military. On Monday, a Chinese foreign ministry spokesman called American calls to slow AI innovation “fearmongering” and part of a Cold War playbook to hobble a rival.

The diplomatic exchange followed remarks Saturday from Anthropic Chief Executive Dario Amodei, who wrote that a Chinese lead in AI would “pose grave danger” for the world and urged continued U.S. chip-export curbs and other measures to slow Chinese progress. Anthropic is preparing for a public listing.

Chinese regulators have pushed domestic companies to match specific U.S. advances. When Anthropic earlier in 2026 released its Mythos system, which can automatically detect security flaws, Beijing responded with alarm and pressed local firms to develop comparable capability, the Journal reported. Last month, Z.AI said it had matched the technology.

Z.AI co-founder Tang Jie told employees in July that the company would spend the next two years focused on what he called the “Touch High Project”—building AI systems capable of improving themselves, the same concept that prompted Jacob Coxon to resign from Anthropic. Tang also said the company would invest in “extreme safety governance,” including billions of dollars to better understand how AI makes decisions, a process currently considered largely opaque. Z.AI and other Chinese AI companies have said their goal is artificial general intelligence—systems that match human cognitive abilities.