An insider perspective on China and AI, from Biden’s NSA | The Cognitive Revolution

Strategic Framework for US-China AI Relations:

• Sullivan advocates for "intensive management" of competition rather than a Cold War, explicitly rejecting a zero-sum "end state" where one side wins.
• The proposed "blueprint" involves competing vigorously while establishing "sufficient guardrails" and maintaining deep diplomatic engagement to prevent conflict spillover.
• Sullivan cites the back-half of the Biden administration as a proof-of-concept for managing this rivalry, stating the approach can withstand the advent of powerful AI.
• He explicitly rejects the concept of a "grand bargain" or strategic condominium with China, viewing intense competition as a chronic condition rather than a solvable issue.

Assessment of AI Timelines and Risk:

• Sullivan identifies the possibility of transformative AI arriving within "the next couple of years" as a distinct planning assumption, though he does not view it as an inevitability by 2027–2030.
• He acknowledges the "AI 2027" scenario (intelligence explosion) as credible but argues policy should focus on concrete, manageable risks (security, economics, society) first to build institutional capacity.
• He prioritizes immediate national security risks—such as cyber/bio threats by non-state actors and "wonder weapons"—over abstract existential "PDoom" scenarios for current policymaking.

Export Controls and the "Small Yard, High Fence" Doctrine:

• The US strategy focuses on denying China access to high-end compute to prevent it from gaining a military and intelligence advantage, without seeking total decoupling.
• Sullivan disputes the effectiveness of the "Japan analogy" regarding oil embargoes, arguing that targeted semiconductor controls are distinct from an existential stranglehold and unlikely to trigger military conflict.
• He explicitly opposes Dario Amodei's suggestion of "convincing" China to give up competing, labeling regime change as an invalid US policy objective.
• A key concern is the "civil-military fusion" doctrine in China, which makes it impossible to restrict chips to "benign" uses, necessitating broad controls on high-end hardware.

Domestic Adoption and Policy Implementation:

• Sullivan notes a significant lag in US government adoption of AI tools, citing bureaucratic inertia and legal constraints as primary obstacles, even within the Pentagon.
• He criticizes the Trump administration's AI Action Plan for a "dissonance" between its stated goal of staying ahead of China and the reality of decisions like the H-20 chip export approval.
• He expresses concern over US dependence on foreign compute, arguing the US should build the "lion's share" of its data centers domestically rather than outsourcing the bulk of training runs to the Gulf or elsewhere.

China's Strategic Challenges and Soft Power:

• Sullivan outlines four specific Chinese threats: "China Shock 2.0" via state-directed overcapacity, coercive influence over speech/internet standards, pre-positioned malware on US critical infrastructure, and the world's largest peacetime military buildup focused on Taiwan.
• He acknowledges China's "soft power" move of promoting open-source AI but argues most nations prioritize US technological leadership ("American tech stack") over open-source availability.
• His base case assumes the consolidation of power under Xi Jinping will continue, though he remains open to people-to-people exchanges despite government restrictions.

Military Applications and Lessons from Conflict:

• Observations from Ukraine suggest a shift toward "abundance" and attritability in warfare, with early integration of AI into drone platforms while maintaining human-in-the-loop protocols.
• Sullivan identifies the US military's inability to adopt AI at speed due to bureaucratic inertia and defense prime contractor constraints as a major risk compared to the PLA.
• He warns that the US must accelerate the integration of AI in intelligence, logistics, and command-and-control to avoid falling behind.

Future Outlook and Stability:

• Sullivan rejects the notion of a stable "arms control" equilibrium similar to the Cold War nuclear standoff, viewing the AI landscape as too uncertain to define fixed rules yet.
• He prefers a policy of "feeling our way across the stones," favoring uncertainty and ambiguity in declarations of "strategic dominance" rather than explicit, rigid frameworks.
• He views the "AI 2027" scenario as a valid possibility but argues that focusing on it now distracts from building the necessary consensus and tools to manage current, tangible disruptions like job displacement and misinformation.