Beijing Think Tank Urges US-China AI Risk Dialogue Before Broader Tech Talks

A proposal from a Ministry of State Security-affiliated institute to build shared AI risk terminology signals that Beijing views definitional alignment as a prerequisite to any formal governance...

A Chinese think tank affiliated with the Ministry of State Security has advised Beijing and Washington to establish common terminology for describing artificial intelligence risks before attempting to resolve their wider technology rivalry, according to reporting by the South China Morning Post.

The recommendation came from the China Institutes of Contemporary International Relations (CICIR), which operates under the Ministry of State Security, China's principal civilian intelligence and security agency. CICIR's institutional position gives its public statements particular weight as a gauge of official thinking, though the institute's views do not constitute formal government policy.

The SCMP report notes the CICIR call came within days of Chinese President Xi Jinping and U.S. President Donald Trump agreeing to formalize some form of AI-related engagement between the two governments. The specific terms of that agreement, including its scope, timeline, and enforcement mechanisms, were not fully disclosed in publicly available records as of September 29, 2026.

CICIR's core argument is that negotiating concrete technology policy agreements is not viable while the two countries lack a shared framework for describing AI-related harms and threat categories. Without common definitions, the institute argues, any governance arrangement is vulnerable to misinterpretation and unilateral reinterpretation by either party.

The proposal reflects a structural challenge that has complicated arms control and technology diplomacy in previous eras. In nuclear negotiations, the United States and Soviet Union spent years establishing agreed definitions for warhead categories, delivery systems, and verification terms before substantive arms limitation treaties became possible. Historians and policy analysts have documented this definitional groundwork as a prerequisite to agreements such as SALT I and the INF Treaty.

AI governance talks face a comparable problem in a more compressed timeframe. Terms including autonomy, dual-use capability, frontier model, and algorithmic weapon carry different technical and legal meanings across regulatory traditions. Without resolution of those definitional gaps, overlapping but incompatible national AI safety frameworks could produce persistent misunderstanding at the policy level.

The CICIR recommendation focuses specifically on risk vocabulary rather than on market access, export controls, or intellectual property, which are the more visible points of friction in the broader U.S.-China technology relationship. This framing positions the proposal as a technical and diplomatic confidence-building measure rather than a direct challenge to existing U.S. Restrictions on semiconductor exports or Chinese AI development.

U.S. Export controls imposed in 2022 and expanded in subsequent years have restricted the sale of advanced graphics processing units and other AI-relevant chips to Chinese entities. Those controls remain in effect. It is not known from publicly available sources whether the Xi-Trump AI engagement agreement addresses export control policy directly.

The SCMP report also noted, separately, that Hong Kong businesses are lagging behind mainland Chinese peers in integrating AI across business functions. A survey conducted by the consultancy Accenture found that only 10 percent of Hong Kong firms had integrated generative AI across multiple business functions, compared with approximately 21 percent of surveyed mainland Chinese firms, according to a senior Accenture executive quoted in the same publication. Accenture attributed Hong Kong's lower rate to legacy information technology infrastructure and a risk-averse corporate culture.

That divergence in AI adoption rates is relevant context for the diplomatic proposal. If the internal Chinese market is already fragmenting along adoption lines between Hong Kong and the mainland, the challenge of building shared international risk frameworks across significantly different regulatory and technological environments is likely to be substantial.

What would clarify CICIR's proposal is a formal response from U.S. Government bodies responsible for AI policy, including the Office of Science and Technology Policy or the State Department's Bureau of Cyberspace and Digital Policy. No such response has been reported in publicly available sources as of the publication date of this article.