In Which Areas of Technical AI Safety Could Geopolitical Rivals Cooperate?

Some AI safety research areas are safer to share internationally than others. The paper identifies AI verification mechanisms and shared protocols as particularly suitable for cooperation between geopolitical rivals, suggesting that not all safety research poses equal national security risks.

Cooperation on AI safety can advance dangerous capabilities if not carefully managed. The authors highlight that the specific technical details being shared matter significantly—some collaborative research could inadvertently help rivals develop more powerful AI systems, which is a key concern for policymakers weighing cooperation benefits against security risks.

Historical precedent shows nations do cooperate on strategic technologies despite rivalry. By examining how countries have managed cooperation on other sensitive technologies, the paper suggests that international AI safety cooperation is not unprecedented and can be structured with appropriate safeguards.

Current US-China AI cooperation demonstrates both the potential and challenges. The paper uses this case study to illustrate real-world tensions between wanting to address shared AI risks and protecting national interests, showing why a nuanced approach to cooperation is needed rather than blanket acceptance or rejection.

Existing risk management frameworks need to be adapted for AI safety specifically. The authors argue that standard approaches to managing sensitive technology cooperation may be insufficient for AI safety research and recommend developing new protocols tailored to the unique characteristics of this field.