How the US and China Can Cooperate to Reduce Urgent AI Risks

In this piece, authors Mark MacCarthy & Carl Schonander (Brookings Institution) discuss the renewed U.S.–China dialogue on artificial intelligence following the recent Trump–Xi meeting, arguing that both countries should pursue a practical, technical exchange on AI risk reduction rather than treat the talks as an arms control negotiation. They contend that building trust through regular communication and information sharing offers a more realistic path to improving AI safety than seeking formal agreements or reciprocal concessions.

The authors recommend focusing discussions on concrete AI safety challenges, including cybersecurity threats, the potential misuse of AI in biological or nuclear contexts, and reliability issues such as hallucinations and defective code. They argue that the dialogue should emphasize testing protocols, red teaming, risk assessment, and mitigation techniques, allowing both countries to exchange best practices without requiring binding commitments or synchronized policy decisions.

MacCarthy and Schonander also caution against allowing broader geopolitical disputes to derail the initiative. They argue that issues such as open-source AI development, access to advanced U.S. models, and export controls should remain separate from technical AI safety discussions, which should instead be led by qualified experts capable of addressing shared risks while avoiding politically charged negotiations.

Looking ahead, the authors suggest that successful bilateral cooperation could eventually evolve into a broader international framework for AI risk mitigation modeled on existing transgovernmental networks. However, they stress that any future collaboration should remain narrowly focused on technical risk management rather than global AI regulation, with the ultimate objective being a shared understanding of critical AI risks and effective methods for reducing them…

CONTINUE READING…