On Which AI Red Lines Do Geopolitical Rivals Agree? Mapping Shared Fears Across the US-China Divide

Background

International AI agreements face a fundamental challenge: why would geopolitical rivals constrain capabilities that might give them an advantage? The standard answer—mutual fear of catastrophic outcomes—only works if rivals actually share concerns about the same capabilities.

This project inverts the usual framing. Instead of asking "what red lines should exist?" we ask "what capabilities do adversary states themselves fear?" If both the US and China consider certain AI capabilities destabilizing, these become natural candidates for international agreement.

There is precedent for this approach. The US and China agreed in November 2024 that humans, not AI, should control nuclear weapons decisions. Both signed the Bletchley Declaration acknowledging potential for serious harm. The IDAIS dialogues have produced joint statements from US and Chinese AI scientists. But systematic analysis of where fears converge and diverge is lacking.

Research Questions

1. What AI capabilities do US and Chinese officials and military doctrine identify as threatening or destabilizing?
2. Where do these concerns overlap? Where do they fundamentally diverge?
3. Which shared concerns are most amenable to international agreement, given technical and political constraints?

Deliverables

A research paper (6,000-10,000 words) that:

• Catalogs US and Chinese concerns about destabilizing AI capabilities (from official statements, doctrine, think tank and academic analysis, PLA writings or state media/Congressional testimony)
• Maps areas of convergence and divergence
• Analyzes which shared concerns could form the basis for international red lines
• Proposes 2-3 specific red line formulations that might achieve bilateral or multilateral support

Scope

Primary focus on US and China as the key axis of AI competition. Secondary consideration of EU, UK, and Russia where relevant. Focus on capabilities both sides might fear regardless of who develops them: autonomous weapons, AI-enabled bioweapons, attacks on critical infrastructure, loss of human control, recursive self-improvement.

Methodology

• Weeks 1-6: Collect and analyze US and Chinese sources
• Weeks 6-9: Comparative analysis identifying convergences and divergences; consult with specialists to validate interpretations
• Weeks 9-11: Develop red line proposals based on shared concerns; assess feasibility given verification and dual-use constraints
• Weeks 11-12: Draft paper and solicit expert feedback

International red lines are only achievable if major powers see them as serving their interests. This project identifies the subset of potential red lines where self-interest aligns across geopolitical rivals.

This advances AI safety by: (1) focusing diplomatic efforts on achievable agreements rather than aspirational ones, (2) providing US policymakers with evidence that red lines serve American interests (countering "but China won't agree" objections), (3) informing Chinese engagement by understanding their genuine concerns rather than projecting Western assumptions, and (4) identifying specific red line formulations with realistic paths to agreement.

Even if comprehensive international agreement proves impossible, understanding shared fears enables confidence-building measures, crisis communication protocols, and narrow agreements on specific capabilities—all of which reduce catastrophic risk.

Mentees are active researchers who will co-author the research paper.

Mentee 1 and 2 (US Analysis): Primary responsibility for collecting and analyzing US sources. Drafts US concerns sections and contributes to comparative analysis.

Mentee 3 and 4 (China Analysis): Primary responsibility for collecting and analyzing Chinese sources. Ideally has Mandarin reading ability or experience with Chinese foreign policy. Drafts China concerns sections and contributes to comparative analysis.

Mentees collaborate on the comparative analysis and red line proposals.

Research skills and ability to analyze primary sources (government documents, policy statements) Familiarity with international relations or security studies concepts Analytical writing ability (2,000+ word sample) For China-focused mentee: Mandarin reading ability strongly preferred, or substantial coursework/experience in Chinese politics and foreign policy

Preference for European business hours availability.