China-U.S. AI to soon converse: a preview of interactions between the two AI superpowers

According to Beating Monitoring, the United States and China are preparing to launch a new official AI safety dialogue, attempting to establish a “guardrail” to prevent crises in their technological competition. As reported by The Wall Street Journal, the U.S. side will be led by Treasury Secretary Scott Bessent, while the Chinese side will include officials such as Vice Minister of Finance Liao Min in preliminary discussions of mechanisms.

This official engagement aims to establish a crisis-management mechanism similar to those during the Cold War. The core issues the two sides plan to discuss include: preventing unpredictable behavior by AI models, controlling autonomous military systems, and addressing the threat of hackers and other non-state actors abusing powerful open-source tools. Industry speculation suggests that both sides may explore establishing a high-level “AI hotline.” Experts familiar with China’s position note that China is open to discussions on technical safeguards and governance, and that the key tone of its participation in the dialogue is “not seeking alliances, but seeking stability” (Stability, not alignment).

However, this dialogue meant to prevent problems before they happen is facing a test from the strongly confrontational posture of the U.S. lead negotiator. According to reports from Bloomberg and several other outside media outlets, Bessent has recently taken an extremely hardline stance on AI issues with China: in mid-April, he publicly claimed that Anthropic’s cutting-edge models would ensure the U.S. suppress China in the field of AI; just last month, he openly criticized a senator who invited Tsinghua University scholars to participate in a U.S. AI safety forum, asserting that “the real threat to AI safety is any country other than the U.S. setting global standards.” This dual-track strategy—seeking to prevent things from getting out of control while also excluding other countries from rule-making—creates the biggest uncertainty for the negotiations.

Historically, this is not the first time the U.S. and China have attempted official AI engagement. The Biden administration facilitated the first round of high-level talks in 2023, but because China assigned officials from its diplomatic system rather than technical experts to lead, substantive technical discussions were limited. However, beyond official channels, a “second track” (non-governmental) communication channel lined up by Henry Kissinger before his death in 2023 is still in operation. Former Microsoft executive Craig Mundie, along with representatives from Tsinghua University and China’s leading large-model companies, continues to maintain private exchanges on the safety and alignment of frontier models.