China frames AI as a national security threat amid US-China rivalry

When China's top intelligence official calls AI "a new arena for strategic rivalry among major powers" ^[1], the stated goal is risk prevention - but the underlying logic runs toward more Party control over data and the internet at home. Chen ties AI risk directly to ideological security, treating AI-generated rumors and disinformation as a lever hostile forces could pull to undermine domestic stability, which reframes tighter data and internet control as a defensive necessity rather than a political choice ^[1]. This is not a new instinct: back in 2023, Chen's earlier article on digital threats already flagged China's dependence on foreign AI and high-end chips as a strategic vulnerability, well before AI became the headline issue it is today ^[7]. Read together, the throughline is consistent - external technological dependence gets recast as an internal control problem, and the security vocabulary of "strategic rivalry" does double duty as a justification for domestic tightening.

Strip away the diplomatic language and both governments are leveling nearly identical charges at each other. China's Ministry of State Security warns that foreign spy agencies could exploit deepfakes for "cognitive warfare" - recruitment efforts, defamation campaigns and disinformation designed to destabilize the country ^[2]- and separately warns that unofficial relay services accessing foreign AI models like Claude and GPT create risk of data leaks and illicit data trading ^[3]. Meanwhile, US agencies accuse Chinese AI firms of the mirror-image offense: systematically extracting data and capability from US frontier models through "aggressive, malicious, and targeted distillation" ^[4], and a prior congressional report went so far as to call DeepSeek a "profound threat" for allegedly routing US user data back to China-linked infrastructure ^[6]. Each side describes the other as the aggressor stealing data and manipulating information, while describing itself purely as the target defending against exactly that behavior.

Unlike the more abstract rhetoric about "cognitive warfare" or regime security, the September 8, 2026 joint advisory from CISA, NSA and the FBI names names and puts a scale on the accusation: six Chinese AI companies - DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI - are accused of extracting billions of tokens from millions of exchanges with US frontier models since late 2024, "likely with Chinese government awareness" ^[5]. That specificity is what turns a background rivalry into an active policy dispute, since it names concrete companies and gives US officials a discrete incident to point to when discussing accountability measures, rather than the more diffuse fear of a strategic AI gap ^[4].

Neither government's framing goes unchallenged by its own public. On the US side, online reaction to hawkish proposals - including one former official's suggestion of considering military strikes on Chinese data centers - has been dominated by pushback accusing the framing of being sensationalized, with commenters noting the underlying report never actually advocated bombing anything. Skepticism also surfaced around the distillation accusations themselves: one widely watched analysis questioned how cleanly "theft" applies here, pointing out that AI training methodology and hardware are fairly standardized, making it murkier than classic industrial espionage to say what exactly is being taken. On the Chinese side, online discussion has drawn a similar contrast, framing Beijing's approach to AI risk as fundamentally a state-control and censorship problem rather than an existential one, unlike the more diffuse, profit-driven oversight debate happening in the US. In both countries, the loudest official warnings are running ahead of what a chunk of the public and independent commentators are willing to accept at face value.

Even where both governments might want a bilateral AI agreement, verification remains the sticking point - analysts note that neither Washington nor Beijing can be certain the other side would honor commitments it can't independently confirm ^[8]. That distrust is compounding a parallel domestic argument in the US over whether to slow down frontier AI development at all: pushback from figures like Donald Trump and Mike Johnson has framed any US pause as effectively handing AI supremacy to China, on the logic that a unilateral slowdown only matters if the other side would actually reciprocate it ^[9]. With that backdrop, the credibility gap on both sides isn't just rhetorical friction - it is the specific obstacle standing between the current cycle of mutual accusation and any workable AI safety or governance arrangement.