The concrete goal of an end to the hostilities in Ukraine has been replaced by a host of vague Russian demands concerning a range of issues and regions.
Alexander Baunov
Source: Getty
Alongside official pushback, Chinese tech and policy communities are meeting calls for a slowdown with skepticism—rather than rejection—that mirrors the reactions in the United States.
Months of news stories about artificial intelligence (AI) models “going rogue” culminated in the leading American AI CEOs calling for slowing the speed of AI development. Proposals for a slowdown—what is often called “pacing the frontier” of AI—sparked heated debate in Silicon Valley and Washington, much of it focused on the implications for AI competition with China.
But what does China think about calls to slow the pace of AI development?
Early answers to this question focused on how China’s Ministry of Foreign Affairs (MFA) and state media outlets responded to Anthropic CEO Dario Amodei’s September essay, “We Must Pace the Frontier.” Those official reactions were harsh and dismissive, calling the essay “fearmongering” and a “Cold War script,” leading some to argue that China had rejected any form of cooperation on safety outright.
Digging deeper into both official reactions and the wider tech policy discourse reveals a more nuanced reality. Instead of an outright rejection of “pacing” on geopolitical grounds, much of the Chinese tech and policy community appears to be meeting calls for a slowdown with skepticism. Many skeptics focused on the motivations of Amodei’s calling for a slowdown, while others focused on the feasibility of the proposal. In conversations, some influential members of China’s AI policy community have expressed growing concern about emerging frontier risks and an interest in pacing as one policy response.
The wider Chinese technical and policy community is in no way embracing pacing. But there’s a meaningful difference between rejection and skepticism, one with significant implications for policy.
The idea that AI development should be either slowed down or paused entirely has existed for years in U.S. AI policy debates. But this summer the proposal moved to the very center of these discussions.
Over the course of June and July, Sam Altman, Anthropic executives, and over 1,000 employees of frontier AI companies all publicly stated that safety may require deliberately slowing the pace of development. These statements received some coverage in mainstream Chinese media but didn’t spark official government statements or major policy debates.
That changed on September 12 with Dario Amodei’s essay that advocated for pacing. The piece laid out the potential risks of unfettered AI development and Amodei’s plan for a potential slowdown in new model development. It also warned that America’s ability to slow down “will be limited by the lead that U.S. companies have over authoritarian regimes, chiefly the Chinese Communist Party.” To maintain American AI labs’ current lead, Amodei specifically called for tightening semiconductor export controls on China and a crackdown on Chinese “distillation” of U.S. models.
Within days China’s state media outlets and its Ministry of Foreign Affairs reacted. When asked about the essay, an MFA spokesperson called it “fearmongering.” The Global Times, a nationalistic state media outlet, was more explicit: “On the surface, this appears to be a ‘voice of reason’ focused on global AI safety. But a closer reading of the full text reveals the ulterior motives at play: the article conceals numerous provisions aimed at containing China, making it essentially a ‘Cold War script’ for the field of AI.”
Subsequent editorials from the Global Times and China Daily, another state-run media outlet targeted at foreign audiences, laid out a variety of attacks on Amodei’s piece. They argued that Amodei’s two-step plan—hold Chinese AI companies back through export controls and then use the resulting lead to build a global safety regime—as a thinly-veiled attempt to exclude China from international AI governance. Both pieces argued that American AI companies and the U.S. national security community rely on smearing China to justify a variety of self-interested goals, such as greater export controls to hold back Chinese competitors, out-of-control capital expenditures, and the construction of regulatory moats against American competitors. The China Daily went further, criticizing American views of AI competition as an “existential, winner-take-all contest.”
While some of these arguments lightly touched on the idea of pacing itself, the pushback overwhelmingly focused on the way that Amodei and others talk about China. This is to be expected. If someone presents both a technology policy proposal and an attack on the Chinese Communist Party, the CCP will invariably focus on the attack, not the proposal.
Reporting and analysis from China’s wider tech and business media ecosystem offers an insight into how pacing is being framed in those circles.
Much of that coverage has been relatively factual: tracking the positions of AI executives on pacing and describing the different camps in U.S. AI policy. China’s media and wider policy ecosystem closely tracks the nuances of international AI policy debates, both to anticipate future developments and to draw out policy mechanisms that could be adapted for China.
The more analytical and opinionated pieces often emerge from WeChat Public Accounts—roughly analogous to Substack posts in the U.S. media ecosystem—some of which are reposted by Chinese mainstream media. Many of these pieces are centered on deep skepticism about the motivations of American AI executives. These discussions often echo or directly draw on arguments from skeptics in the United States. Two common threads focus on pacing as rooted in corporate self-interest: either to build regulatory moats around their businesses, or as a distraction from real issues in their business fundamentals during the run-up to IPOs.
Several financial media outlets have focused on AI companies’ IPOs, in particular, Sam Altman’s announcement in September that OpenAI would delay its listing due to concerns over safety. One analysis from Phoenix New Media said Altman was using safety as a “more convincing excuse” for delay than the real reason: weak financials. It leveled the same accusation against Anthropic, calling its public messaging a “deft shift in the capital markets narrative.”
Anthropic’s subsequent release of its prospectus for its public listing quieted some of the criticism around its financial status. But much of the media coverage about the letter remained cynical about the company’s motivations. One in-depth analysis of the company’s trajectory ran under the headline, “With the Trillion-dollar Unicorn Going Public, How Much of Anthropic’s Idealism Remains?” It criticized what it described as steadily eroding commitments to safety in the face of mounting financial pressures. “[Anthropic] has never stopped talking about safety, and it has genuinely paid a price for it. But the prospectus’ language also makes clear that when safety ideals collide with commercial growth, the ideals have to give way. In the business world, there is no Garden of Eden.”
But alongside these more narrow, cynical reads, some influential commentators had a more nuanced read on Amodei’s essay and pacing debates more broadly. Xiao Qian, a former diplomat and Vice Dean of the Institute for AI International Governance at Tsinghua University, analyzed the proposals through the lens of the Trump-Xi summit and accompanying AI-focused dialogue. In an article in the South China Morning Post, Xiao catalogued the many open questions that would need to be answered before any international agreement could be reached: Does the pause apply just to training runs or to certain forms of deployment? How would compliance be verified and who would do it? “Until these questions are answered, it is not realistic to expect a definite answer from any government to the larger question of whether it would agree to slowdown.”
Just two days after Amodei’s essay, an essay by a former editor for influential state media outlets, She Zongming, constructed an elaborate analogy to the plot of a classic Chinese martial arts saga. In the story, endless battles were fought over obtaining a dangerous book that holds the most powerful secrets of martial arts mastery—The Nine Yin Manual. To put an end to the chaos, the greatest martial artist of the age, Wang Chongyang, called together the most capable martial artists to battle once and for all, and the winner takes the manual. Wang wins. Rather than using the manual to dominate his peers or sharing it with others, Wang chooses to lock the text away so that no one obtains such power. After Wang’s death, the secrets of the manual slowly leaked out into the broader community, causing some good and some harm.
The core argument of the piece is contained in the headline: “AI is The Nine Yin Manual, but Dario is not Wang Chongyang.” It gives serious credence to the threats posed by advanced AI, while also arguing that Amodei, and likely any American AI leader, cannot be the one to safeguard the secrets of advanced AI. In the story, Wang was able to play the role he did because both his technical prowess and moral character were widely respected, he didn’t make money from selling the coveted object, and all parties acknowledged him as an impartial arbiter. While Amodei doesn’t meet these criteria, the piece argues, when other American AI leaders also shared the same concerns over the speed of AI development, pacing becomes a worthy discussion. Rather than trying to completely hide away this modern day Nine Yin Manual, the world should set limits to its use, and encouraging third-party evaluators is a right step toward guarding these limits.
In conversation, members of China’s technical and AI policy community partly map onto these narratives in media coverage, but they also diverge in places. The majority of policy attention remains focused on promoting AI applications across the Chinese economy. Understanding of what “pacing” means is uneven in Chinese AI circles, and the idea has generally been met with skepticism about motivations and tractability.
One common thread in these conversations is to put the onus for pacing on the U.S. AI companies. This reflects a certain logic: If China’s frontier AI models lag six to twelve months behind the leading American models, then China should have six to twelve months of advance notice of emerging risks and doesn’t need to slow down immediately. While versions of that logic lead some to wave away discussion of pacing, others view recent rogue AI incidents as a warning of what China will likely face in the months and years ahead.
Some influential figures in Chinese circles express an interest in pacing, but there remains a sense that publicly calling for a slowdown would be a bold posture. This closely mirrors private western discussions of the “AI pause” letter of 2023, which garnered many notable signatories but went unsigned by some influential figures who agreed with the proposal, but deemed it too far outside the mainstream of policy debates. Events of the past year significantly shifted these dynamics in the United States. It remains to be seen if the same evolution will happen in China.
The way that pacing caught the attention of China’s AI policy community meant it would struggle to receive serious consideration. Timing and the messenger matter, and with Amodei’s essay coming out shortly before the Trump-Xi summit in Washington, China’s official reaction to the essay was swift and harsh. The wider media and policy discourse on pacing was largely skeptical of the proposal as well, but not uniformly so.
For the time being, the idea of China joining the United States in a form of AI pacing—one that the U.S. government itself remains hostile to—is far-fetched. The pace of AI development and proliferation of risks has rapidly shifted the discourse within the U.S. frontier labs, bringing pacing squarely within the corporate Overton window. Whether it will have a similar effect in China remains to be seen.
The authors used AI to translate quotes for a first draft. All translations were then human reviewed and edited.
Senior Fellow, Asia Program, and Co-Director, China AI Initiative
Matt Sheehan is a senior fellow at the Carnegie Endowment for International Peace, where his research focuses on global technology issues, with a specialization in China’s artificial intelligence ecosystem. He is also co-director for the China AI Initiative.
Sophie Zhuang
Research Assistant, Asia Program
Sophie Zhuang is a research assistant in the Carnegie Asia Program.
Carnegie does not take institutional positions on public policy issues; the views represented herein are those of the author(s) and do not necessarily reflect the views of Carnegie, its staff, or its trustees.
The concrete goal of an end to the hostilities in Ukraine has been replaced by a host of vague Russian demands concerning a range of issues and regions.
Alexander Baunov
In an increasingly competitive landscape, Seoul should expand its current partnership with Washington and deepen ties with Brussels and New Delhi.
Darcie Draudt-Véjares, Milo McBride
Too many national interests, too many EU institutions, and too little political will are preventing Europe from acting in its own best interests.
Judy Dempsey
A once-unlikely partnership between President Prabowo and former president Joko Widodo has ended, with important consequences for Southeast Asia’s largest democracy.
Sana Jaffrey
Does the NPP have the will or vision to create a new social contract for Sri Lanka?
Bhavani Fonseka