A Trump-Xi Jinping Hotline Is Not a Substitute for Global AI Governance

We should not wait for an AI catastrophe to create the institution capable of investigating it

Opinion

By Yiannis Damellos

September 21, 2026

Let me just tell you what I think: a U.S.-China notification mechanism is useful as a safety layer, but it should not be mistaken for an international AI safety regime. The current proposal disclosed by Treasury Secretary Scott Bessent is explicitly about the two countries notifying each other about AI incidents that could threaten national security. However, AI risk is not inherently bilateral. An AI system developed in the United States or China can affect people, infrastructure, financial systems, cybersecurity, and other countries without those countries having any seat at the table. The two governments therefore have an obvious reason to communicate, just as nuclear powers did, but communication between two competitors does not create global accountability.

What does? There is already something remarkably close to the institutional architecture that we urgently need right now. In 2025, the U.N. created an Independent International Scientific Panel on AI, consisting of 40 experts from all five U.N. regions. Its mandate is explicitly scientific and independent, and the panel is intended to function as an early-warning and evidence mechanism. Separately, the U.N. established a Global Dialogue on AI Governance involving all member states as well as academia, industry, civil society, and the technical community.

I could easily say, “Let the U.N. handle it.” But it's not enough. What we really need is a four-layer system that primarily addresses prevention by setting international safety standards that frontier AI developers must meet before deploying systems capable of causing serious harm. Secondly, we need a detection and early-warning mechanism: an independent scientific body capable of identifying emerging risks and requiring rapid disclosure of significant incidents. Then, we need not only bilateral but also multilateral channels — including a U.S.-China hotline — so that governments can communicate immediately when something goes wrong. Finally, we must enforce accountability. If an AI company deploys a system that violates internationally established safety requirements, responsibility should not disappear into the familiar corporate fog of “the technology behaved unexpectedly.” Regulation needs identifiable obligations, independent auditing, mandatory incident reporting, and consequences for deliberate or negligent violations.

This is not to say that notification itself cannot be preventive in some circumstances. If Washington and Beijing know immediately that an AI system has behaved dangerously, they may be able to stop escalation, isolate infrastructure, or prevent another actor from misinterpreting the event. Reuters reports that the proposed dialogue is already considering dangers, including uncontrollable AI agents and cyber threats from non-state actors. So the proposed mechanism is not useless. It's necessary, indeed, but radically insufficient.

At the end of the day, we should not wait for an AI catastrophe to create the institution capable of investigating it.

The U.N. scientific panel I mentioned earlier states that current safeguards are not keeping pace with AI capabilities and describes its role as an early-warning mechanism. Maybe we should give it a try because the alternative generates a deeper problem by leaving the architecture primarily to Washington and Beijing: both governments have incentives that can conflict with safety. They are competing for technological, economic, and military advantage. That doesn't mean either side will deliberately endanger humanity; it means that neither should be the sole judge of how much risk is acceptable for everyone else. On the other hand, a U.N. governance structure should be able to separate scientific assessment, regulation, corporate compliance, and geopolitical crisis communication, providing security to all of us.

The mistake is not that Washington wants an AI hotline with Beijing. Both are scared. The mistake would be believing that a hotline between the world's two leading AI powers is a substitute for global AI governance, and given that the U.N. has already established the scientific panel and global dialogue, it is grounded in an institution that actually exists.