Analysis

China is becoming increasingly dependent on AI. New technologies are now central to plans for economic growth, ensuring the party’s rule, pushing Chinese technology and companies overseas, and monitoring dissent. At the same time, CCP leaders fear that AI threatens the party’s rule. The CCP’s concerns that emerging technologies will erode its monopoly on power means that AI occupies a paradoxical position in China as both a panacea for the country’s problems and a potentially existential threat to the regime's survival. 

This two-sided meaning of AI means that the more that AI becomes a perceived threat to the CCP, the more the CCP will likely employ the same technologies to try to mitigate such threats. In other words, the risk that AI exceeds the party’s control means that the CCP will likely try to expand its use of AI mechanisms to try to prevent such outcomes, which in turn feeds back into the development of new AI technologies to address the party’s demands for more powerful AI tools. 

This also means that China is increasingly hooked on AI and the need to rapidly innovate ahead of US competitors. Beijing therefore opposes recent calls from leaders in the US tech sector to slow down AI development (see “Business Matters” below). The CCP believes that if it decelerates China’s AI work, China will fall behind the accelerating progress of US-based companies and more broadly fall behind the US in overall economic and political competition. Safety and security be damned, the conflict with the United States is existential. Economics and technology are central to that competition, and AI is central to economics and technology.

This logic is pertinent ahead of General Secretary Xi Jinping’s planned visit to Washington, where potential cooperation on AI safety will likely dominate discussions. If AI is such a threat to CCP rule, then that should push the party towards cooperation and global regulation. But the uncertainty generated by such threats also incentivizes the use of AI to attempt to use the same technologies to quell challenges to the party’s rule. In short, the CCP’s fears for its grasp on power will continue to dominate how the party simultaneously thinks about AI as both the key to its survival and threat to its existence. 

US negotiators (and tech leaders) will therefore face an uphill battle if they hope to ensure cooperation with the CCP on AI development. Instead, the CCP will likely push for faster innovation at home all while disingenuously professing China’s dedication to safety and cooperation.

Analysis

This week’s planned summit between President Trump and General Secretary Xi is unlikely to improve relations between the US and China. The summit did catalyze a series of bilateral side meetings with Marco Rubio and Wang Yi, and Scott Bessent and He Lifeng (see “Business Matters” below), which largely built on the nonsubstantive pleasantries exchanged in previous talks. The challenge for both sides is that if the bar for success is simply meeting, regardless of whether the meeting results in desired change, then we face a period of managed conflict  rather than meaningful cooperation. 

For both Washington and Beijing, the incentive to cooperate is unclear, largely because their ultimate goals are unclear. The Department of Commerce, for example, is pushing to “close loopholes” for US exporters to China in sensitive technology areas, although chip sales continue. At the same time, President Trump’s call to “unleash” AI faces calls from both within industry and the administration to approach the technologies with greater caution. Mixed signals are also prevalent from China’s side. The CCP, for example, is pushing the Chinese tech sector to develop at full speed, but not if that contravenes the CCP’s demand for party control over business needs, as companies like Manus found out earlier this year. 

Collectively, if neither side is sure what it wants out of the relationship, then it is challenging to define success beyond just showing up. Both sides will look for a headline, which will likely come in the form of a questionable agreement to cooperate on AI safety (see “Business Matters” below). This means that US businesses hoping for a change in policy direction (such as loosening of tariffs, greater access to rare earths, an end to PRC sponsored hacking and IP theft, or a relaxation of restrictions on Chinese open source models) will be disappointed. Instead, the current phase of frosty relations and mutual suspicion will continue. Businesses should not expect any breakthroughs.

Analysis

In the lead-up to the late-September Trump-Xi meeting, US Treasury Secretary, Scott Bessent, and Chinese Vice Premier, He Lifeng, met for a second time to lay the groundwork for negotiations. The agenda of their discussions expanded, though, from focusing on tariffs and advanced tech sales to also including the possibility of AI regulation. 

As the longest-standing concern, trade war and tariffs now appear to be routine and transactional. Chinese Ministry of Commerce officials confirmed they are negotiating with their US counterparts to develop a reciprocal tariff reduction framework valued at US$30B, although the details remain closely guarded. The proposed tariff reductions are a prerequisite for establishing the Board of Trade that was proposed during the US-China trade negotiations in May of this year. If negotiations succeed, the new tariff framework and all future US-China trade concerns would be routed through the Board, potentially streamlining what has otherwise been an unpredictable, difficult, lengthy and ad-hoc process.

AI regulation was the newest item added to the agenda for Bessent and He’s meeting, as is the possibility of an AI executives meeting on the side of the formal summit. While Bessent indicated the US is open to discussing shared AI risk with China, including the future of Chinese open-weight models in US markets, the reality is far more complex. President Trump is keen on maintaining America’s lead over China and has already announced his opposition to regulating domestic AI technologies; the closest thing to regulation at the moment is Washington’s framework for monitoring domestic, closed-weight models. At the same time, US AI companies want to retain access to cheaper Chinese open-weight models to help accelerate innovation. 

As discussed above in “From Zhongnanhai,” China is torn about how to handle AI. On the one hand, China’s AI industry is one of the primary growth drivers among an otherwise deeply troubled economy, and Huawei just announced ten new chipsets for AI infrastructure that will further close China’s technology gap with its US competitors. On the other hand, the head of China’s Ministry of State Security, Chen Yixin 陈一新, published an article encouraging heavier regulations for AI, especially frontier models, because of the risk these models pose to national security (see “Espionage Alert” below). The result is that China’s economic future now appears to depend on a technology that poses an existential threat to Party rule. It also means that, for as much as China complains that it cannot access advanced US chips and AI models, it has also now implicitly acknowledged that doing so would threaten US national security.

Given the ambivalent attitudes and policies of both governments toward AI, it is not clear how the two sides could agree on a definition of shared AI risk, let alone specific steps to mitigate it. Both sides are aware of the threat that AI poses, but neither party wants to be the first to apply the brakes and risk the other side gaining the upper hand. Chinese and US security experts have both proposed a “nuclear-style safeguard,” but the nuclear analogy does not go far. When compared with the challenges of monitoring and verifying AI development, counting nuclear warheads and delivery systems, as well as estimating their effects, are simple tasks. 

With so many potential stumbling blocks, no agreement on AI is likely. The second order concern then becomes how Washington plans to insulate other agenda items from a potential failure to agree on AI.

Analysis

The CCP, the government of China, and China’s AI companies hate Anthropic, whose CEO, Dario Amodei, is currently at the center of US-China tensions. The CCP rejects his calls for an AI “slowdown” and instead advocates for a China-centric view of global AI safety. In particular, Amodei’s statement that “a Chinese lead in AI would pose grave danger for the United States and the world” and his calls to continue to restrict chip and other technology sales to the PRC sent a jolt of electricity through a multitude of nerves in Beijing. 

The CCP has never been adept at accepting criticism. One-party dictatorships never are. The CCP took particular offense at Amodei’s criticism because Amodei showed the world that China pays lip service to the need for “global AI governance” (albeit on China’s terms), while driving full speed toward AI dominance. 

The CCP doth protest too much. Distillation attacks against US frontier labs, continued cyber attacks against US critical infrastructure, and widespread IP theft from US companies are documented and frequent. Beijing objects to Amodei’s criticisms not because they are unfounded, but because they are valid, and more to the point because Amodei’s words carry weight with the congress, elements of the administration, and other governments and companies around the world. Calling China a threat undermines Beijing’s attempts to portray China as a cooperative partner that champions AI safety. If China loses global partners (and customers), then that damages not only Chinese companies’ bottom lines, but also China’s global political clout. 

It is not uncommon for private companies to find themselves subjects and objects of geopolitical tensions. But the CCP’s attention towards Anthropic and other US-based AI companies speaks to a deeper paranoia about how these technologies could threaten China’s national security, its ability to outcompete the United States, and to the CCP’s rule at home (see Espionage Alert below). Amodei is not unique in his description of China’s AI threats: Sam Altman endorsed Amodei’s letter, Alex Karp of Palantir and Alexandr Wang at Meta have both highlighted the Pentagon’s lack of spending on AI compared to China, and other industry leaders have similarly echoed the need for coordination on AI development to combat China. 

This conflict represents nothing new. Anthropic was already a target of CCP espionage and sabotage; that won’t change. The CCP reaction to the comments does highlight clearly that the CCP sees US tech companies and their leaders as threats to China’s power and security, and even CCP legitimacy. Tens of thousands of other US companies in AI, aerospace, energy, new materials, robotics, biotech, and other high tech industries are also CCP targets. This is the war the United States is already in; and the economy is the battlefield. 

Unlike Amodei, most CEOs don’t realize it and if they do, they do not take sufficient action. This public verbal conflict between Amodei and the CCP can be useful if it spurs US tech firms to take steps required to protect themselves against CCP aggression. Amodei understands that geopolitical competition is now an inseparable part of his business; CEOs who fail to understand this may soon find themselves out of business.

Analysis

China’s reluctance to decelerate AI development stems in part from a fear that technology can undermine the CCP’s power. Speaking publicly, the head of China’s Ministry of State Security (MSS), the party’s intelligence bureaucracy, expressed nervousness about what AI could mean for China’s (or more explicitly the CCP’s) security. AI has the potential to fundamentally sever the party’s capacity to rule by eroding its monopoly on the information, social monitoring, employment, and economic performance that gives the party control over society. Fear of this outcome is heightening the party’s attempt to not only fuse its priorities within AI companies and the technology’s development, but also to drive the use of AI in aggressive espionage or surveillance methods to try to prevent others (both domestically and abroad) from using similar technologies against China. Moreover, the Party is concerned that even without explicit human agency, the technology itself could turn on the CCP.

This fear will increase the CCP’s use of AI to monitor internal dissent, persecute detractors around the world, and to attempt to cripple China’s adversaries, including the United States. Chinese AI-enabled espionage attempts will heighten. This makes for a somewhat schizophrenic messaging on AI from the CCP, whereby it emphasizes “AI safety,” which itself is a euphemism for ensuring CCP control over the technologies, while engaging in ever-intensifying AI-enabled espionage and even cyber attacks. 

This bifurcated messaging is apparent in the recent unveiling that the MSS, despite public caution about AI’s threats to party rule, continues to use the same technologies to conduct operations. Recent reporting by Anthropic, for example, noted how PRC government-aligned operations “used Claude to track, profile, and recruit Uyghurs and Uyghur armed formations in Syria,” in addition to other religious groups that the CCP deems threatening. While Anthropic’s reporting notes these explicit cases, the expansive use of AI in espionage does not stop at so-called dissident communities. Rather, the MSS will rely on such tools to monitor not only PRC citizens, but anyone that it deems a threat to the CCP. That category is capacious, and includes all democratic governments and all critics worldwide. The CCP, even when compared to other autocracies like Russia and Iran, is notoriously thinskinned. 

As discussed above in “From Zhongnanhai,” the pervasive fear motivating both the building and curtailing of similar technologies makes for mixed signals from the party about AI. It arguably disincentivizes cooperation on AI safety for technologies that the party thinks it needs to quash opposition, including the use of AI-enabled espionage against selected targets. US companies need to be aware that such espionage attempts,including techniques outlined in Anthropic’s reporting, will increase in the months ahead.