The US and China are quietly preparing to launch a new official AI safety dialogue, aiming to build guardrails for what is shaping up to be the defining technology race of the decade. According to The Wall Street Journal, the US side will be led by Treasury Secretary Scott Bessent, while China has tapped Vice Minister of Finance Liao Min for preliminary discussions.
This would be a kind of Cold War-style crisis management mechanism for artificial intelligence. The proposed agenda includes preventing unexpected AI model behavior, controlling autonomous military systems, and dealing with non-state actors — like hackers — who might abuse powerful open-source tools. Industry insiders speculate the two sides could even set up a high-level "AI hotline." Experts familiar with China's position say Beijing is open to talking about technical safeguards and governance, with a guiding principle described as "Stability, not alignment."
But this whole exercise in preemptive guardrail-building faces a major obstacle: the lead US negotiator's combative posture. According to Bloomberg and other outlets, Bessent has been unusually aggressive on China AI issues. In mid-April, he loudly proclaimed that Anthropic's frontier models would ensure US dominance over China in AI. Then, just last month, he publicly attacked a senator who invited a Tsinghua University scholar to a US AI safety forum, declaring that "the real threat to AI safety is letting any country other than the United States set global standards." This two-track strategy — seeking to prevent runaway AI while excluding others from rulemaking — is the wild card that could derail the talks.
This isn't the first time the two countries have tried official AI engagement. The Biden administration managed to kick off a high-level dialogue in 2023, but it was hamstrung because China sent foreign ministry officials rather than technical experts, limiting real substance. However, outside official channels, a track-two (unofficial) dialogue — quietly established by the late Henry Kissinger before his death in 2023 — remains active. Former Microsoft executive Craig Mundie has been holding private discussions with representatives from Tsinghua University and China's leading large model companies, focusing on frontier model safety and alignment.