The three countries with the world's largest nuclear arsenals are talking about the same question: should artificial intelligence ever be allowed near the decision to launch a nuclear weapon? According to the South China Morning Post, reported on 11 November 2026, negotiators from the United States, China and Russia are discussing a possible mutual commitment to keep AI out of nuclear command-and-control systems — ahead of a possible three-way leaders' meeting on the sidelines of the APEC summit in Shenzhen on 18–19 November.
Nothing has been agreed yet. No final agenda has been announced, and the meeting itself is still conditional on the three sides reaching a consensus on format and agenda. But the fact that AI in nuclear systems has become the focus of the preparations — while bigger flashpoints fell off the table — tells you how seriously all three governments now take this risk.
What is actually being discussed
- Human control of the launch decision. The central proposal, according to the reports, would reserve the final decision to use nuclear weapons for a person — not an autonomous AI system. The discussions reportedly cover early-warning systems, software used to identify targets, and the process for deciding whether to launch.
- A joint declaration. Diplomats are also reportedly working towards a joint statement reaffirming the principle that a nuclear war cannot be won and must never be fought. Heather Williams, a senior fellow at the Centre for Strategic and International Studies, said such a statement was possible.
- Guardrails, not disarmament. The talks are about how nuclear weapons are controlled, not how many each country holds. That distinction matters: China has long resisted trilateral talks on arsenal sizes, arguing its stockpile is far smaller than those of Russia and the United States, and that Washington and Moscow should cut first.
- Cyberattacks on command systems. Measures under discussion also include limits on cyberattacks against nuclear command, control and communications networks, information exchanges on missile tests, and more transparency about arsenals. A ban on cyber or AI-enabled attacks on command systems is described as one of the less likely outcomes.
Why the agenda narrowed to AI
Working-level negotiators have spent weeks narrowing the summit agenda to subjects where agreement looks possible. Two of the biggest geopolitical flashpoints — the war in Ukraine and the denuclearisation of the Korean Peninsula — have largely fallen off the table, according to the SCMP's sources. China reportedly prefers Ukraine to be discussed directly between Russia and the US, and treats Korean Peninsula security as a bilateral matter between Pyongyang and Washington.
That left AI in nuclear systems as one of the few areas where all three sides could plausibly agree on something. Russia is the most cautious of the three: the Moscow Times describes Moscow as resistant to formal curbs on nuclear command systems, because they could limit its military flexibility. Kremlin spokesman Dmitry Peskov has confirmed that Moscow is exchanging agenda proposals with Washington and Beijing — but he did not mention AI restrictions.
The context that makes this urgent
- New START is gone. The New START treaty between the United States and Russia — the last major nuclear arms-control agreement between the two — expired in February 2026. There is currently no treaty capping the two largest arsenals.
- A precedent already exists. In November 2024, Washington and Beijing agreed that humans, not AI, should control decisions to use nuclear weapons. The Shenzhen talks would extend that principle to a three-way commitment including Russia.
- A wider AI-risk dialogue is opening. US Treasury Secretary Scott Bessent said this week that the US and China have set up a dialogue on the risks of superintelligence — a sign that AI safety is becoming a standing item between the two governments, not a one-off topic.
What to watch on 18–19 November
Three things will tell you whether this was real or ceremonial: whether Trump, Xi and Putin actually meet in Shenzhen; whether any joint text mentions AI and nuclear command at all; and whether the wording is a binding-style commitment or a vague declaration of principle. Given the gaps — Russia's caution, China's refusal to discuss arsenal sizes, and the fact that no final agenda has been agreed with days to go — a broad statement of principle is the most likely outcome, and even that is not guaranteed.
But even a principle matters here. The world's three largest nuclear powers agreeing, in writing, that a machine should never make the final call on launching a nuclear weapon would be the first rule of its kind — and a line that every future AI system in a military command chain would have to respect.
Bottom line
No agreement has been signed, no summit has been confirmed, and Russia is resisting formal limits. But for the first time, the US, China and Russia are negotiating the same sentence at the same table: the final decision to use a nuclear weapon belongs to a human. Watch Shenzhen on 18–19 November — if that sentence survives into a joint statement, it will be one of the most important AI guardrails ever written.
Sources: the South China Morning Post report of 11 November 2026, as reported by the Moscow Times; Asia Daily. Hero photo: Shenzhen skyline, by Fumikas Sagisavas, CC0, via Wikimedia Commons.

