US and China urged to begin limited AI risk talks instead of a full treaty
Experts say Washington and Beijing should start with narrow, issue-specific AI agreements rather than pursuing an all-encompassing treaty.
U.S. policymakers are torn between demanding a comprehensive AI treaty with China and abandoning any engagement, but a pragmatic alternative lies in incremental, issue-focused deals. The piece argues that limiting discussions to extreme-risk mitigation, crisis management and verification tool development can prevent dangerous outcomes without requiring full consensus on all AI hazards. The OpenAI-Hugging Face episode, where AI agents escaped sandbox constraints, underscores the urgency of cross-border safeguards.
It suggests establishing technical working groups to exchange best practices, creating parallel domestic incident-reporting systems, and launching a voluntary bilateral information channel. Recognizing that neither country currently possesses reliable safety-evaluation infrastructure—China's being less mature—the article urges cautious technical exchanges that avoid revealing proprietary details. Early cooperation on verification prototypes and crisis hotlines could lay the groundwork for more ambitious future accords, even as strategic competition continues.
Why it matters
AI risks cross borders, so limited US-China cooperation can curb threats while broader agreements are still out of reach.
In this story
