US-China AI Race: Two Scenarios for 2028 Global Leadership
Summary
- • Policy paper outlines two 2028 scenarios hinging on US export control action
- • Compute access — not talent — argued as the decisive constraint on Chinese AI
- • Paper contends CCP already uses AI for censorship, hacking, and military buildup
- • Authors argue America must tighten loopholes now or risk ceding AI norm-setting to authoritarians
Details
Paper calls for tighter US export controls on AI chips to preserve compute lead over China
Authors argue current controls have been effective but contain exploitable loopholes. The CCP has used those gaps to maintain near-frontier model capabilities despite chip restrictions. The paper urges policymakers to close loopholes proactively rather than reactively.
Authors project transformative AI systems will arrive by 2028, making the current window decisive
The 2028 timeline is the paper's central organizing assumption. The argument is that whoever shapes norms before that threshold will define how powerful AI is governed globally — making near-term policy choices disproportionately consequential.
CCP identified as already using AI for censorship, dissident repression, government hacking, and PLA modernization
The paper does not treat AI-enabled authoritarianism as a future risk — it frames it as a present reality. These characterizations reflect the authors' framing in an advocacy document, not independently verified claims.
Authors argue distillation attacks — not chip smuggling — are the primary vector Chinese labs use to extract American AI innovations
Large-scale distillation involves training a model on outputs from a more capable model, effectively transferring capabilities without accessing the underlying weights or training compute. The paper treats this as a significant and underaddressed threat to US AI advantage, distinct from hardware export control evasion.
Paper acknowledges Chinese labs have world-class talent; compute constraints alone limit their frontier progress
This framing separates the CCP-as-adversary argument from any claim about Chinese researcher capability. It also implies that if compute constraints erode — through loopholes, distillation, or future domestic chip production — talent gaps would not serve as a compensating brake.
Scenario 1 (US acts) is described as the condition under which AI safety engagement with China becomes possible
The paper argues democratic AI leadership and US-China safety cooperation are complements, not alternatives. From a position of strength, the US is better positioned to negotiate norms and safety frameworks — an argument aimed at safety-focused audiences who might be skeptical of hardline export control policy.
Paper frames the core task as avoiding self-inflicted disadvantage, not outpacing China through new investment alone
Authors characterize America as approaching competition 'from a position of great strength,' arguing the primary risk is policy inaction rather than any fundamental deficit in US AI capability — an optimistic framing designed to make the policy ask seem achievable.
Policy = regulatory/government action, Strategy = positioning argument, Security Alert = active threat claim (per authors), Insight = analytical argument from authors, Context = background framing
What This Means
This paper — an advocacy document, not neutral analysis — makes the case that the US-China AI race will effectively be decided in the next two years, and that compute access is the decisive lever. For AI professionals and policy audiences, the argument that distillation attacks represent a major strategic vulnerability is notable and likely to surface in regulatory debates around model API access and output licensing. If the paper's 2028 timeline for transformative AI gains traction in policy circles, it could accelerate bipartisan pressure for tighter export controls and new rules around how American labs expose their models to external queries. The framing also sets up a direct tension between open AI development norms — which facilitate distillation — and national security interests, a conflict the field will increasingly have to navigate.