Analysis

The CCP’s announcements about AI are rapidly shifting from a position of unimpeded, enthusiastic embrace to enforced restriction and even curtailment. At least part of the reason for this seemingly sudden policy reversal is because the CCP AI and tech innovation more broadly create a dilemma for the CCP. On the one hand, China’s leaders hope to rely on their tech companies as drivers of global innovation, while on the other, exerting absolute control over emerging technologies, a policy that stifles the innovation they seek. 

This dilemma has led to several prominent contradictions. First, the party is exploiting its reach into China’s economy and society to spur tech innovation through systemic investment, incentivized distillation, and enforced model usage. At the same time, the party intends to rely on these same mechanisms to demand full oversight over new technologies and their alignment with the party’s priorities. In effect, the CCP is bargaining that its early support for tech companies grants it control over their technologies. Conversely, those same tech companies are bargaining their involuntary Faustian pact with the CCP will grant them success even if they must accede to the party’s paranoid demand. 

Second, the CCP aims to balance its desire for control against its strategy of pushing Chinese open source overseas, particularly in the Global South, to build dependence on Chinese models at the expense of the United States’ closed models. Open models, by their very nature, contradict the party’s insistence that Chinese models are ultimately a function of China’s national and economic security. Increasing discussion that the CCP might also restrict open models for foreign users suggest that the party’s dilemma is shifting towards the need for political control. For companies around the world trying to embrace AI in the workplace, this is particularly concerning. Companies built using US or Chinese models, as recently demonstrated by the US government’s restrictions on Anthropic’s Fable 5, will likely need to consider the strategic risks of losing access to these models. 

Additional emerging evidence points to the CCP moving increasingly towards closing off some foreign access to Chinese technology in favor of maintaining autocratic control. The CCP is imposing sweeping, system-wide restrictions on foreign engagements through export controls on dual-use technologies, curbs on outbound tech investments, preference for closed over open models, limits overseas travel for some tech professionals, and intensifying demands for party membership, All this collectively suggests that the CCP increasingly sees foreign engagement as inherently threatening to party control. It is likely that each time the United States or China makes a move regarding AI, the other side will likely respond in kind: Export controls beget export controls and so on. We may well therefore see a technological future in which the US and China, despite their otherwise myriad differences, converge over their treatment of AI and create separate global AI blocs.

Analysis

According to the US government, AI frontier models are now integral to national security. Recent restrictions against Anthropic, OpenAI, and Google illustrate the shift in how US-China competition is forcing an increasingly tight partnership between US based AI companies and the government. AI companies complain that, to date, regulation is ad hoc and until the government develops clear, consistent, and predictable guidelines, its regulation will hamper US competitiveness and restrict model development.  

The intentions of these restrictions appear to be leaning towards sector regulation while the US government builds the expertise to evaluate the technologies’ capacities, rather than cede questions of national security to the private sector. This is not  a misplaced intention. AI models, particularly, if left unregulated, can cause untold damage. The challenge is that directives as to which models should be  restricted, blocked, or outright banned appear to react to the latest corporate announcements, rather than to follow a clear policy objective and strategy. Conditions that tech companies curtail access for non-US citizens, for example, highlights a condition for what counts as a point of contention for the newly expanded vision of national security that will be almost impossible for Silicon Valley firms to implement. Much like in China, demands for security therefore appear to be outweighing policies that spur innovation. One such policy is relying on the best people, regardless of citizenship, to outcompete foreign counterparts. 

US government restrictions currently appear to depend on the individual company in question, rather than to apply sector wide. US companies, particularly those working on frontier models, therefore need constant engagement with government contacts to try to preempt or reduce the scope of future restrictions. This lack of a consistent, sector-wide policy also means that other countries and companies are considering the potential strategic challenges of relying on US (or PRC) models, particularly, though not exclusively, for issues related to national security. If building a company on Claude’s latest model, for example, risks losing access to that model, then some users may decide Claude is too risky and instead opt for local alternative technologies. 

Collectively, both the US and China are attempting to navigate a rapidly evolving new technological and industrial landscape as AI diffuses throughout the global economy. An ad hoc regulatory approach is therefore partially understandable as a response to the barrage of new model announcements that are simultaneously enticing and concerning, depending on one’s perspective. The problem is that individual companies now face the brunt of determining what and how regulations will continue to direct emerging technologies in potentially market-defining ways.

Analysis

As we hit the midpoint between official meetings of presidents Trump and Xi (the next is anticipated in late September on the sidelines of the UN General Assembly meeting), tensions are growing alongside nominally conciliatory efforts. While China has resumed purchases of US soybeans and Boeing secured a $3.62B deal with China Southern Airlines, Beijing has ramped up offensive measures to poach talent, steal advanced tech, and impinge on free trade.

China’s efforts to woo top technology staff are not new, but one effort in particular is now paying off. Even Realities Technology is a Chinese-funded, Shenzhen-headquartered, smart-glasses maker founded by former Apple employee Will Wang. Funded largely by Meituan and Tencent, it raised $150M in the pre-Series B round, giving it a valuation of $1B and unicorn status. Despite being a Chinese company, the majority of the company’s engineers reside in the US and the bulk of its sales target the US and European markets. This is, therefore, a case where the loss of US-based personnel to China has resulted in the creation of a competitor that appears to have benefited from, if not heavily borrowed, US companies’ R&D. China has long sponsored talent recruitment programs, as well as creating a new visa fast-track for STEM talent last fall. Companies would be wise to review top talent’s terms of employment and ensure working conditions encourage their retention. It is important to note that Apple has not, to this point, claimed that Wang took any Apple IP with him to his new company.

Beijing has also again tightened its grip on rare earths while potentially circumventing US export controls. New restrictions prevent Chinese companies from selling specific, in-demand rare earths to ten US companies, including MP Materials and USA Rare Earth. Under the claim that the metals and products containing them are dual-use products, that is, possessing both civilian and military applications, China is preventing the two largest US rare earth companies from purchasing necessary magnets and metals used to create semiconductors and other advanced tech.

At the same time, US Commerce Secretary, Howard Lutnick, suggested that Beijing has obtained one of Dutch company ASML’s extreme ultraviolet (EUV) lithography machines, the sale of which to China the US has banned since 2019. These machines are key to producing the frontier chips and would significantly accelerate China’s chip development. ASML denies Lutnick’s claim, saying that all machines are carefully logged and accounted for, and Dutch officials have pushed back against aggressive US intervention into Dutch businesses.

In short, while China makes symbolic gestures to attempt to placate and deceive US officials, it continues aggressively to pursue a multipronged strategy for outcompeting US companies, upending supply chains, and obtaining banned technologies. It also means that companies need to see China for what it is. It is not a reliable partner or even merely a risky investment, but a predatory public-private conglomerate that sees trade and business as zero sum contests. Business for China is therefore war by other means.

Analysis

A growing number of PRC-based startups are being founded by former Silicon Valley employees. This is an increasingly common career trajectory within the tech industry and should worry US companies. The concern is not entirely because of the risk of IP theft, although as recent cases between Apple and OpenAI demonstrate, that is a persistent concern among tech companies that consistently hire each other’s employees. Rather, the concern is that China’s systemic support for tech startups is drawing vital tech talent away from the United States and towards China. This support includes steering financial services, to subsidizing R&D through university partnerships, and assistance with finding local staff.

One recent example is that of a former Apple employee, Will Wang, who raised $150m in the pre-Series B round from Chinese financiers for his Shenzhen-based smart-glasses firm, Even Realities Technologies. The company aims to challenge Meta’s market dominance in the smart glasses market and was recently valued at $1b, making it a unicorn. Wang’s previous employment at Apple’s wearable tech divisions does not mean that he stole IP to found his venture, but rather that the skills and know-how gained on the job introduced him to how US-based competitors conduct business that he was then able to translate into his own venture. Wang’s story is far from unique: Shunyu Yao, the researcher behind some of the tech industry’s foundational work on AI agents, left OpenAI to become Tencent’s Chief AI Scientist in December 2025. 

This is a pipeline that is increasingly enticing to PRC citizens in Silicon Valley. Facing increased suspicions of loyalty in the United States, including being banned from working on some frontier models or technologies, not to mention the vast financial push by the PRC government to attract “returnee” talent (and the draw of being closer to family), many Silicon Valley employees are starting to see China as an attractive alternative. This financial push in particular is startling: the CCP’s theoretical mouthpiece recently proclaimed that China needs to recruit entire “innovation teams,” rather than individuals, to build entire systems for technology entrepreneurship. Talent programs, funding offers, and employment or founder opportunities are booming in China, which means that Wang and Yao’s careers are likely to become a more common story. 

The US tech sector faces a double challenge. Restrictions from the US government curb companies’ abilities to hire the necessary foreign expertise to build the frontier models that keep them ahead of the competition. At the same time, US companies now face additional risks that the same experts may leave with an intimate understanding of corporate practices and then form overseas competitors that could threaten these companies’ market share. This means that US companies will increasingly need to consider how to retain their people, ensure protections for existing employees, and enforce “need-to-know” firewalling to avoid leaking sensitive information as employees depart the company. As OpenAI’s case against Apple demonstrates, this is an issue that is far from unique to China. But unlike Apple and AI, Chinese companies lay beyond the jurisdiction of the US legal system, and the Chinese government is aggressively supporting startup ventures. 

Analysis

This week, more than 1,400 guests and representatives from over 1,100 international companies will participate in the 2026 World AI Conference (WAIC) in Shanghai. This is a four-day (17-20 July) event co-organized by China's Ministry of Foreign Affairs (MFA), the National Development and Reform Commission (NDRC), and four other PRC ministries alongside the Shanghai municipal government. Although this gathering may appear to be a world-class trade show, in actuality, WAIC has been built and staffed by the PRC party-state to bring together the specific ministries, state labs, universities, and technology firms that represent the CCP’s preferred instruments for stealing trade secrets and finding sources for acquiring additional sensitive information.  In short, this conference is an intelligence operation.

This is not a fringe interpretation. Last week, China scholar Matt Brazil noted the PRC’s "whole-of-system" approach to competition, which integrates political control, cyber operations, intelligence collection, influence activities, tech acquisition, and other elements of the party, state, and business community to increase Beijing’s comprehensive national power. This whole-of-system approach allows the CCP to selectively activate specific institutions most useful to its objectives, including government ministries, state-owned enterprises (SOEs), universities, research institutes, and other "non-traditional collectors," including so-called private companies. WAIC's official structure illuminates this strategy, featuring PRC ministries as co-organizers, state key labs as providers of technical infrastructure, prominent PRC universities as suppliers of talent and legitimacy, and an entire conference pillar dedicated to identifying and recruiting individuals with useful expertise. (Read “expertise” not just as brain power, but specific, confidential, protected knowledge.)

Recruitment of key personnel is a focal point of this year’s WAIC 2026. The "Talent and Recruitment" track runs alongside the conference's exhibition and forum activities, representing a large-scale, in-person PRC head-hunting operation for AI expertise and potential sources of commercial intelligence. Even the conference's official app facilitates this objective: "Hi WAIC" was launched by Donghao Lansheng, a PRC state-owned enterprise established in 2013 under the direct jurisdiction of the Shanghai Municipal People's Government. The app is built on a "multi-agent architecture" that integrates “event matchmaking,” "circle-based social networking," and think-tank data, meaning that foreign attendees who register through it are, in effect, handing their personal and professional data to a platform built and operated by the Chinese government.

None of this is new. Last month, we highlighted the Five Eyes intelligence alliance warning that Chinese intelligence services use platforms like LinkedIn to target individuals with access to sensitive government, defense, and commercial information. WAIC compresses that same playbook into four days and one location: instead of a cold call on social media, foreign engineers, researchers, and executives will receive a badge, an app, a booth conversation, and a business card, all inside a legitimate, state-sanctioned setting designed to lower one’s guard. As Brazil's framework makes clear, the participating institutions (ministries, state labs, universities, and SOEs like Donghao Lansheng) are not merely neutral participants in an open marketplace of ideas; they are the same institutions the party-state selectively activates when it pursues intelligence collection. WAIC places these institutions in direct, sustained contact with exactly the foreign talent Beijing has indicated that it needs to target.

This does not mean that US companies should avoid WAIC. The conference remains a legitimate window into China's AI network, and skipping it entirely cedes ground. However, it does mean treating attendance as a counterintelligence challenge, and not just a business trip. Companies should brief staff before departure on how to recognize and handle intelligence approaches, and instruct them to report such contacts to their security office. Think carefully before bringing your personal or your company’s phones or laptops, and avoid sharing anything beyond the minimum required information WAIC will showcase genuine PRC AI progress, but it will also feature the CCP's preferred tools for espionage.