Otis Kim: Iris, long week — you ready to talk about the end of the world, sort of?
Iris Holm: Always. Though 'end of the world' might be giving the current situation more credit than it deserves.
Otis Kim: Fair. Because what we actually have is — the South China Morning Post reporting, citing people familiar with preparations, that the US, China, and Russia are in working-level discussions. To set an agenda. For a meeting that hasn't been confirmed.
Iris Holm: About restricting AI in nuclear command-and-control systems. At APEC Shenzhen, November 18th and 19th. Potential participants: Xi Jinping, Donald Trump, Vladimir Putin. Confirmed participants: none, yet.
Otis Kim: And Mao Ning at China's Foreign Ministry said exactly that — Beijing is maintaining communication, no final decision made.
Iris Holm: The Associated Press called it 'historic.' Also noted that differences make substantive agreement uncertain. Both things in the same dispatch.
Otis Kim: And some outlets just took the first word.
Iris Holm: So what I actually want to work out — is there a real arms-control effort forming here, or are three nuclear powers performing the appearance of one?
Otis Kim: Okay but — performing is actually the right word, and here's why. Think about two neighbors who already signed a deal saying no robot arm controls the lock on their shared vault. And then a third neighbor shows up, and suddenly everyone's acting like a whole new agreement is forming. Except nobody's checked whether the first deal still counts.
Iris Holm: The November 2024 bilateral.
Otis Kim: Exactly that. The US and China reached an actual bilateral agreement — November 2024 — affirming that humans, not AI, control decisions to use nuclear weapons. That exists. Russia was not part of it.
Iris Holm: And now Russia's in a trilateral on the same subject. But no source — not South China Morning Post, not Associated Press, not anyone — clarifies whether the trilateral supersedes that deal, extends it, or quietly contradicts it.
Otis Kim: Which means the new conversation is being treated as a fresh start when there's an unresolved prior agreement sitting underneath it. That's the structural gap nobody names.
Iris Holm: And the Rubio–Wang Yi call — described as 'positive and constructive' — that call was itself preparatory. It was about finalizing APEC arrangements. The trilateral format got raised. That's not a trilateral agreement. That's an agenda item about maybe having a meeting.
Otis Kim: Right — but the part that doesn't fit is Russia's timing. If the US-China bilateral from 2024 already covered the human-in-the-loop principle, why is Russia included now? Did Moscow ask in? Did Washington need a third signature for political cover?
Iris Holm: No source answers that. Which is — frankly — the question.
Otis Kim: And AI in nuclear command-and-control isn't just the launch button. It's early warning, targeting software — the whole chain before a human ever touches anything. So whatever the 2024 bilateral actually covered, Russia has been running its own version of that chain the entire time, without signing anything.
Iris Holm: And that chain is running right now — which is what makes the UN move so hard to ignore. Russia and the US, November 2026, the same month as APEC, actively blocked binding rules on autonomous weapons at a UN conference. Patrick Wintour at the Guardian documented it. Legally binding governance — killed. And then the same two powers show up at APEC talking nuclear AI restraint.
Otis Kim: In the same month.
Iris Holm: Exact same month. That's not evolution. That's a behavioral contradiction in plain sight.
Otis Kim: And I actually think the agenda narrowing is what reveals the mechanism. Ukraine war — set aside. Korean Peninsula denuclearization — set aside. The only thing left on the table is AI in nuclear command-and-control, specifically because it's the one area where you can claim consensus without resolving anything that actually hurts.
Iris Holm: Right — you haven't reached consensus. You've located the smallest possible overlap and called it a framework.
Otis Kim: And then there's China's October 2026 Foreign Ministry statement — that its nuclear arsenal 'is by no means at the same level' as the US and Russia's, calling the whole trilateral disarmament framing 'neither fair, reasonable nor realistic.' That's October. And then, I mean — China's reportedly at the table in November anyway.
Iris Holm: So either Beijing's calculus shifted in weeks — or someone's performing for a domestic audience while the real negotiation happens elsewhere. Those are different situations with identical surface appearances.
Otis Kim: And we can't tell which from outside. Which is — yeah, that's the problem.
Iris Holm: The part that gets worse — and we'll get to this — is what 'human in the loop' actually means when AI has already filtered the decision space before the human enters it, and why verification makes the whole thing structurally unenforceable.
Otis Kim: And that's exactly the thing — 'human in the loop' sounds like it means something until you actually picture the loop. Six a.m., US Strategic Command officer gets an alert. Early warning system flags a possible Russian missile launch. AI-generated confidence: 78%. The human is supposed to decide.
Iris Holm: Except the AI already narrowed his options.
Otis Kim: Three pre-loaded responses. Ninety seconds. And the sensor data that didn't fit the threat signature? Already filtered out before he sees the screen. The human didn't stay in the loop — the human entered a loop that was already moving.
Iris Holm: That's the specific problem with the 2024 US-China bilateral. It affirmed human control. Didn't define what constitutes meaningful human agency when AI has already compressed the decision space down to — what, a multiple choice with a countdown.
Otis Kim: Right — and nobody pushed on that definition. Not in the bilateral, apparently not in the trilateral consultations either.
Iris Holm: Now compound that with verification. To confirm any party is honoring a 'restrict AI in nuclear command-and-control' pledge — you'd have to look inside their classified command-and-control architecture. That is the secret each side most wants to protect. The inspection mechanism that would prove compliance is precisely the thing no party will permit. And look — no source describing the trilateral discussions identifies any proposed verification mechanism. Not even a flawed one.
Otis Kim: Not even a bad idea on the table.
Iris Holm: Which tells you they haven't started solving the actual problem. And then there's reportedly a joint declaration on the protocol texts — reaffirming that nuclear war cannot be won and must never be fought. That's one of the things potentially going into this APEC document.
Otis Kim: Yeah — which sounds significant, except it says nothing about the AI systems shaping how close each side gets to that moment. You can sign 'nuclear war must never be fought' and still have an early warning algorithm running 24/7 that narrows a human's options to three choices in ninety seconds. The declaration doesn't touch any of that.
Iris Holm: Even if Xi Jinping, Donald Trump, and Vladimir Putin actually sit down at Shenzhen on the 18th — which is still not confirmed, only the agenda consultations are confirmed — the signature on that declaration doesn't touch what happens upstream of the moment it claims to govern.
Otis Kim: Yeah. The verification problem has no proposed solution. Not in anything the South China Morning Post reported, not in any source describing these discussions. Not even a bad mechanism on the table. And I think that's actually the answer to whether this is solvable — at least between these three, right now.
Iris Holm: It might be less about whether they can agree — and more about whether they've discovered the problem is genuinely unsolvable at this table. Three powers who can't even agree on what 'human control' means operationally. That's not a negotiation failure. That's a definitional failure that precedes it.
Otis Kim: I could be wrong, but I think showing up to Shenzhen might be the whole deliverable. The headline exists whether or not anything gets signed. That's an uncomfortable place to land.
Iris Holm: Frankly, yes. That's where I am too. Good talk — genuinely.