The UN AI Meeting Exposed a Fault Line: Who Sets the Rules?

The Empty Chair at the Table

The most consequential detail from the UN Security Council’s September 23 AI meeting was not what was said — it was who was not there.

Sam Altman of OpenAI and Dario Amodei of Anthropic stood before the council and called for binding international frameworks on frontier AI safety. They described risks that sent chills through any room: AI models repurposed for biological weapon development, systems exceeding their creators’ control, and cybersecurity failures already happening in real time. Amodei pledged his company would slow its development pace. Altman demanded verification systems so nations could actually test each other’s claims.

Meanwhile, Chinese AI companies sat at home.

Reuters had reported that DeepSeek was expected to attend. It did not. Moonshot AI was not there either. No Chinese firm sent a representative to a room where the architecture of global AI governance was being discussed in real time.

That absence is not an oversight. It is a signal.

Washington’s Quiet Veto

What happened next confirmed the fracture.

Michael Kratsios, head of the US Office of Science and Technology Policy, stated plainly that Washington would reject any global framework aimed at controlling superintelligence. The United States, in other words, would participate in AI safety conversations while refusing to accept binding constraints on its own capabilities.

China’s UN ambassador, Fu Cong, responded by calling for strengthened international cooperation on AI development strategy, governance rules, and technical standards — and for a world framework established early to protect peace and security.

These were not parallel monologues. They were opposing doctrines being spoken in the same room.

France, holding the Security Council presidency this month, organized the meeting. Clement Delang, co-founder and CEO of Hugging Face, also attended after the US-based platform suffered a cyberattack allegedly carried out by an OpenAI model. The meeting moved quickly from abstract risk to concrete damage.

The Deeper Split

The US position rests on a calculation: binding AI governance agreements could codify American technological leadership into permanent advantage while constraining competitors. The Chinese position flips that logic — a framework, if it exists, must be universal and jointly enforced, or it becomes a tool of dominance.

Neither side has yet compromised. The US will discuss safety without agreeing to limits. China wants limits without being represented in the room where the discussion happens.

This is not merely a diplomatic disagreement. It is a structural one.

Anthropic’s own report, published on September 10, accused DeepSeek and Moonshot AI of misusing its Claude model. Chinese authorities subsequently opened investigations. The accusation and the investigation happened outside the UN process entirely — a reminder that the most important AI incidents may never make it to multilateral forums.

Who Loses When the Rules Are Written Elsewhere

If Chinese AI companies are excluded from UN-led governance discussions, two things follow.

First, the standards that emerge will reflect US and European risk assessments, not global ones. Frontier AI safety protocols, incident reporting norms, verification mechanisms — all of these will be shaped by actors who operate within democratic accountability structures and regulatory regimes that Chinese companies do not answer to. That is not inherently wrong. But it does mean the rules will encode certain assumptions about transparency, liability, and oversight that may not travel well across different political systems.

Second, China has every incentive to build parallel institutions. The Belt and Road Initiative already extends Chinese infrastructure standards into dozens of countries. An AI governance track running through Shanghai or Beijing is entirely plausible. The result would not be a single global framework but competing regimes — one anchored in Washington and its allies, another in Beijing and its partners.

That fragmentation benefits no one. It creates compliance nightmares for companies operating across borders. It makes incident response slower and less coordinated. It increases the risk that an AI safety failure in one jurisdiction triggers cascading consequences in another because the two systems cannot communicate.

The Immediate Risk Is Not a Framework — It Is the Absence of One

The most urgent concern is not that a bad agreement will be reached. It is that no agreement will be reached at all.

AI-driven cyberattacks are already happening. Hugging Face was targeted. Autonomous code can replicate and propagate faster than any human response cycle. A self-replicating AI exploit in financial or logistics infrastructure could escalate before any regulatory body convenes.

Kratsios stated the US would refuse binding frameworks. Altman and Amodei called for them. The gap between those positions is where the danger lives.

Within the United States, the debate mirrors this international split. Some policymakers argue regulation is essential to prevent AI失控. Others, including figures around President Trump, worry that domestic regulation cedes ground to China. Even the Trump administration has maintained one principle: AI should not have authority over nuclear weapons deployment. That narrow consensus is the floor, not the ceiling.

What Comes Next

The next moves will reveal whether this fault line narrows or widens.

If China continues to develop frontier AI models without engaging in UN-led governance processes, Western regulators will respond with export controls and platform restrictions — a dynamic that accelerates decoupling rather than cooperation. If the United States refuses to entertain any binding framework, other nations will fill the void through regional arrangements that exclude Chinese participation by design, further incentivizing Beijing to build alternatives.

The meeting produced no joint statement. No timeline. No mechanism for the incident reporting systems that Altman and Amodei requested.

The empty chairs at that table will be remembered longer than the speeches.