What did China’s State Security Minister Chen Yixin warn about the threats AI poses to China’s political, institutional, ideological, and technological security—including adversaries’ use of deepfakes, AI generated content, social media bots, “public opinion wars,” and “cognitive

Chen Yixin’s warning treated AI principally as a threat to regime security and China’s technological sovereignty—not simply as a generic safety or extinction-risk problem. He argued that hostile actors could use AI to destabilize China politically and ideologically, penetrate its data and infrastructure, and lock it into an unequal foreign-controlled technology order. 2

• Political and ideological security: Chen warned that deepfakes, AI-generated text and images, and automated social-media accounts could fabricate political rumors, spread harmful information, amplify divisions, and support “public-opinion” and “cognitive” warfare—thereby threatening China’s political, institutional, and ideological security. 2

• Espionage and data security: He warned that AI’s ability to aggregate, analyze, and profile massive data sets could aid foreign intelligence collection, including acquisition of state secrets, commercial or trade information, and personal data; he also linked advanced AI to risks to critical information infrastructure. 23

• Technological sovereignty: Chen portrayed AI as a strategic-competition arena in which leading countries may invoke national security to impose technology restrictions, dominate standards, and build closed ecosystems. The policy implication is indigenous capacity, state control of data, algorithms and infrastructure, and resistance to external technological dependence. 2

How this differs from Amodei, Altman, and Musk

• Amodei’s frame is capability control and broadly shared catastrophic-risk prevention. He called for slowing frontier-model capability gains enough for safety measures to catch up, beginning with embedded independent evaluators with meaningful access, then coordination among democratic governments and companies on common standards, and eventually global coordination that includes authoritarian states. 1

• Chen’s frame is sovereignty and political order. The core danger is not merely that advanced models might be unsafe for everyone, but that adversaries could use them to undermine CCP rule, extract Chinese data, and preserve a Western technological advantage. Thus, “safety” also means information control, security screening, domestic control over key technology, and resisting foreign rule-setting. 2

• They overlap on loss of human control. China has also flagged increasingly autonomous AI and possible loss of control as a future safety risk, so there is a real, if narrow, basis for US–China discussion. But that overlap does not erase the larger conflict over access to chips, models, standards, and who sets governance rules. 3

• Altman and Musk broadly sided with Amodei in this episode. Altman said OpenAI should “pace the frontier” and indicated support for third-party safety evaluators; Musk’s public response was simply, “Dario is right.” Their stated focus was therefore slowing and evaluating frontier capability progress, not Chen’s state-centric program of regime protection and technological autonomy. 1

China’s official and state-media reaction illustrates the divide: it cast Western calls to slow AI as fear-driven, competitively toxic, or an effort to constrain China’s technological progress, rather than as neutral safety policy. 2 The planned first bilateral US–China AI-safety engagement therefore reflects a limited shared concern—systems becoming difficult to supervise—within a much wider strategic contest, not a settled common approach to AI governance. 3