U.S.-China AI talks stall as scholars pursue informal safety cooperation
Scholars keep dialogue alive despite stalled U.S.-China AI talks proposing emergency hotlines, shared testing standards and joint safety measures to curb risks.
The formal diplomatic track for U.S.-China AI talks has largely stalled, but researchers and think tanks on both sides have continued to meet informally to address shared safety concerns. These academic exchanges have generated concrete proposals — from emergency hotlines for accidents to common testing standards for biohazard synthesis — even as official negotiations remain frozen. The persistence of these conversations underscores a pragmatic belief among many experts that technical cooperation can reduce risks even when political relations are strained.
Informal Scholar Exchanges Continue
Scholars from leading universities and policy institutes in Washington, Beijing and other global hubs report frequent bilateral meetings held at conferences and through joint workshops. Participants described the sessions as vigorous and productive, producing a range of cooperative proposals intended to reduce immediate safety risks. These gatherings are often hosted by neutral institutions and provide a channel for candid technical discussion away from the headline-level diplomacy.
Proposed Cooperative Measures
Among the most discussed ideas are an emergency hotline to alert partners in the event of an AI-related accident and shared standards to test whether AI systems can synthesize biological hazards. Experts have also outlined protocols for joint red-team exercises, cross-border incident reporting and mutually agreed benchmarks for model evaluation. Proponents argue that such practical measures can create fast, implementable steps that improve safety without requiring a broad political settlement.
Political Pressures Shape Dialogue
Despite the technical traction, political dynamics feed deep mistrust on both sides and shape which proposals gain traction. Many Chinese security scholars view U.S. safety initiatives skeptically, interpreting them as potential levers to constrain China’s technological development. That skepticism is mirrored by American officials who worry Beijing may accelerate capabilities in secret, creating a climate in which cooperation is constantly pressured by broader strategic rivalry.
U.S. Concerns Over Accelerating Capabilities
On the American side, a significant contingent of scholars and policymakers are concerned about the possibility of a rapid, covert push toward advanced AI capabilities, including the theoretical prospect of artificial superintelligence. Beijing, by contrast, has publicly emphasized applied and commercial uses of AI, focusing on immediate industrial and societal benefits rather than public messaging about superintelligence. This divergence in public framing contributes to differing threat perceptions and complicates efforts to build shared risk assessments.
Terminology and Technical Gaps Hinder Consensus
Even when experts agree on the need for cooperation, they often lack a common vocabulary to describe core risks. Long-running collaborative efforts between institutions such as leading think tanks and top universities have produced glossaries precisely because participants could not agree on basic definitions like “loss of control.” Those terminological gaps make it harder to draft mutually intelligible standards and mean that technical working groups spend substantial time clarifying terms before moving to concrete policy options.
Trust-Building Measures and Pathways Forward
Many participants argue that confidence-building steps — small, verifiable cooperative projects with clear technical milestones — offer the most realistic pathway to broader cooperation. These could include jointly administered model-evaluation labs, agreed protocols for disclosure of dangerous capabilities, and third-party audits conducted by international research consortia. Supporters say such measures would allow governments to monitor compliance and build familiarity between scientific communities without immediately resolving larger geopolitical disputes.
Formal diplomacy on AI may remain paused, but the persistence of these expert-level exchanges suggests a parallel track that could be scaled up if trust can be nudged toward reciprocity. Technical proposals developed in academic settings could be pilot-tested and then presented to ministries for consideration, creating a stepwise mechanism from scholar recommendations to government action. For now, informal cooperation functions as a pragmatic hedge against worst-case scenarios and a reminder that shared technical risks can prompt cross-border collaboration even amid strategic rivalry.
As talks at the governmental level wait on political signals, researchers on both sides continue to refine practical proposals and test collaborative models that could later form the basis of more formal agreements. The coming months will show whether these informal measures can translate into durable, verifiable mechanisms that reduce immediate risks while building the trust necessary for deeper U.S.-China AI cooperation.