Transcript | Samm Sacks and Paul Triolo on WAIC 2026, Xi's AI Speech, and Kimi K3
Transcript (courtesy of the fantastic CadreScripts) further down the page. Image by Keya Zhou. Listen in the embedded player above!
This week on Sinica, a rare treat: an in-person recording from Beijing with two dear friends who happen to be two of the very best in the business on technology and China — Samm Sacks and Paul Triolo, fresh off the exhibition floor of the World Artificial Intelligence Conference in Shanghai. We dig into Xi Jinping’s first in-person WAIC appearance and his most extensive statement on AI to date, the launch of the World AI Cooperation Organization, Moonshot’s release of Kimi K3, the Trump administration’s reported push to shut Chinese open-weight models out of the U.S. market, the coming age of agents, the untranslatable problem of ānquán, and what to expect from the first U.S.-China AI dialogue in September.
8:51 – The view from the floor: heat, humidity, robot boxing grandmas, WeChat-gated free water, and the “AI+” vibe — why WAIC 2026 felt less like an AI conference than a sector-by-sector snapshot of China’s entire economy being supercharged with AI, with attendance swelling to some 200,000 tickets
16:32 – Why Xi showed up: what the leader’s first in-person WAIC appearance and his most extensive AI statement to date signal, and why domestic drivers matter as much as geopolitics
18:01 – Chapter and verse: which phrases from the speech will be put to work in the system — “secure and orderly development” and the governance of agents, and Xi’s strikingly extensive language on AI safety after China was frozen out of the Paris process
22:26 – The ānquán problem: one word meaning both “safety” and “security,” the three buckets of AI risk, and how China’s safety community has moved from bias and deepfakes toward CBRN and loss-of-control concerns — Black Mirror versus Star Trek
28:14 – Shanghai’s baby: how WAIC’s ownership structure differs from the CAC-run World Internet Conference in Wuzhen, Chen Jining’s very visible host duties, and whether the center of gravity in AI policy is shifting to the Yangtze River Delta
30:05 – WAICO: what the new World AI Cooperation Organization with its 29 founding members is actually for, Xi’s concrete deliverables for the Global South — 5,000 AI training slots, regional cooperation centers, the MAZU early-warning system — and healthy skepticism about follow-through
34:39 – Kimi K3: what’s technically significant in Moonshot’s big new model, why it’s the fourth arguably frontier-class Chinese release in a single month, the two-way traffic in distillation accusations, and what it all says about the state of the frontier gap four years into export controls
40:33 – Washington reacts: the reported menu of options for shutting Chinese open-weight models out of the U.S. — entity listings, a draft executive order, supply-chain security authorities — and why none of the tools actually fit the problem
49:36 – Strange bedfellows: David Sacks versus the “closed lab duopoly,” the FUD strategy, why some 80% of Andreessen Horowitz portfolio companies reportedly run on Chinese open models, and how gating U.S. frontier models while Chinese weights flow freely supercharges the AI sovereignty argument worldwide
54:55 – The model is infrastructure, the agent is the product: the ByteDance–ZTE agentic phone, the CAC’s new initiative on agent trust and interoperability, and why agents fused into operating systems upend both super-app walled gardens and China’s data protection regime
1:02:27 – An exegesis of kěkòng: the many meanings of “controllable,” the long history of ānquán kěkòng in Chinese tech policy, and the unanswered question of who — CAC, NDRC, or somebody new — actually owns AI safety in either system
1:08:54 – The road to September: what to expect from the first U.S.-China AI dialogue, why Mythos tops the Chinese grievance list, the securitization feedback loop that starves trust-and-safety advocates of resources on both sides, and why recursive self-improvement makes this feel like a last, best chance
Paying It Forward
Paul nominates Tony Peng, whose Substack RecodeChinaAI offers sharp, well-written analysis of the application side of China’s AI industry — part of an impressive new generation of independent China tech writers.
Samm gives a shout-out to Professor Zhu Yue of Tongji University Law School, published in Science and doing pioneering work at the intersection of disability law and AI law.
Recommendations
Samm: The Land and Its People by David Sedaris — laugh-out-loud funny, especially the “Enough is Enough” chapter; Transcription by Ben Lerner, a perfect small novel about fathers, sons, memory, and technology as enabler or disabler of connection; and Didion and Babitz, on Joan Didion and Eve Babitz and the 1970s California rock scene.
Paul: The Party’s Interests Come First by Joseph Torigian — dense but beautifully written, and essential for understanding the current Chinese leadership.
Kaiser: A fiction-only summer! Stoner by John Williams, a small life told most grandly in some of the most beautiful sentence-level writing anywhere; Gilead by Marilynne Robinson, an epistolary novel dense with distilled wisdom from a dying Iowa minister; and Wang Xiaobo’s The Golden Age (黄金时代) in Yan Yan’s excellent new translation — bawdy, hilarious, and super Beijing-y despite its Cultural Revolution setting.
Transcript
Kaiser Kuo: Welcome to the Sinica Podcast, a weekly discussion of current affairs in China. In this program, we will get books, ideas, new research, intellectual currents, and cultural trends in China. Join me each week for in-depth conversations that shed more light and bring less heat to how we think and talk about China.
I’m Kaiser Kuo, coming to you this week from Beijing. As always, a quick reminder before we begin, Sinica is powered by your subscriptions. If you like the work that I’m doing with this show and with the writers and podcasts in the Sinica Network, the very best way to support my work is to head over to sinicapodcast.com and become a paying subscriber.
Subscribers get the complete show transcript, essays and reviews, the full range of offerings from our network, and more importantly, the warm glow of knowing you’re keeping independent, non-hysterical conversations about China alive. So, do it. You will feel great. So, I’ve been really looking forward to this show.
This week I am joined not across 12 time zones, but in person, in the actual flesh, sort of around an actual table — Actually spread out a little more than that — But right here in Beijing in the same room by two people who are not only among the very best in the business when it comes to technology and China, but whom I’m lucky enough to count as genuinely good friends, Samm Sacks and Paul Triolo. Both have been on the show multiple times, but this is still an absolute treat, and the timing could hardly be better.
They have just come up on a plane in one case, on a high-speed rail from another, from Shanghai, where they attended the World Artificial Intelligence Conference, WAIC. Fresh off the exhibition floor, lanyards barely off, opinions fully formed, but minds, of course, still very much open. Folks, it just doesn’t get any better than this. And there is a lot to talk about.
The headline out of Shanghai, of course, was that Xi Jinping himself made his first in-person appearance at WAIC and delivered what is, by some distance, the most extensive public statement on artificial intelligence from him to date. A few things jumped out at me. He opened remarkably, I thought, by actually crediting the United States as the birthplace of the field, Dartmouth University, the summer of 1956. That is not nothing, given the state of the relationship. He announced that the World AI Cooperation Organization, confusingly also WAICO, I think we’re pronouncing it WAICO, I don’t know, is now a reality launched in Shanghai with 29 founding member states alongside some concrete offers to the global south, which I thought were very important as well.
On open source, Xi Jinping said precisely one line that China should encourage open source, which, as our good friend, Matt Sheehan, notes in his excellent annotated read of the speech, I will link to that on the podcast page, restates the status quo rather than staking out anything new. And then there was the closing, which has been rattling around in my head ever since, a classical injunction to adapt as circumstances change followed by a call to make governance ever more precise and to “constantly refine measures to forestall loss of control.” We’ll be looking more closely at that speech and at that phrase in the conversation that follows. One more piece of news from the week deserves mention.
That, of course, is that Moonshot AI released Kimi K3, a big, big, big mixture of experts model with a million-token context window that by Moonshot’s own benchmarks approaches the leading American frontier models on a number of coding and agentic tasks without beating them across the board. And we’ll dig into what all that means. Anyway, with a full open weights promised by the 27th of this month, so around the time you’ll be hearing this. I’m going to leave the gap has closed, hyperventilating to other people, because that’s what the rest of the damn internet is for.
But this is a genuinely notable release, I’ll say that much. We will dig into what it does and doesn’t tell us and look a bit at the response that the Trump administration is now weighing, the different responses, possibly trying to shut Chinese open-weight models out of the U.S. market entirely. Let me introduce my guests properly, though most of you will need no introductions to either of them.
Samm Sacks was until recently a senior fellow at Yale Law School’s Paul Tsai China Center and at New America. And she is now, drumroll please, non-resident senior fellow at the ACF, the Center for America, China, and the Future of Global Affairs at Johns Hopkins School of Advanced International Studies, SAIS, with, of course, friend of the pod, Jessica Chen Weiss, who is just, you know, my lodestar.
Simply one of the sharpest people anywhere on China’s cyber, on data, on AI regulatory architecture, on cross-border data flows, and on the politics of what she’s called managed interdependence. She is at work on a book about data and trust in an age of strategic rivalry, and she has been a friend of mine and of the pod for many, many, many years.
Welcome back, Samm. Great to see you.
Samm Sacks: It is so nice to see you in Beijing, Kaiser. We’ve come a long way from my parents’ porch in Chapel Hill, North Carolina.
Kaiser: Indeed, we have. Indeed. Your folks are still there, right? I mean, I’ll be back. I mean, I missed the place already, but I’m sure I’ll see them.
Samm: Let us know when you’re back.
Kaiser: I will. Paul Triolo is a partner at DGA Albright Stonebridge Group, where he leads the technology policy practice, and he’s an honorary senior fellow at the Asia Society Policy Institute, working alongside my buddy Lizzi Lee, cranking out some really great stuff there. Nobody bridges the full stack better from fabs to high-bandwidth memory and interconnects, all the way up to export control policy and Track II diplomacy, which is what happens when you put an engineer’s training into 30 years or something like that in government service all into one analyst.
He writes the excellent Substack newsletter, AIStackDecrypted, which you should subscribe to after you subscribe to and pay for mine, of course. But Paul is also a dear, dear friend, a repeat offender on this program. His main offense, of course, being his tendency to plunge heedlessly into the technical and technological weeds without pausing to gloss all the terms that he should. Paul, consider yourself officially warned. You are enjoined to avoid doing that this time. No need, by the way, to dumb it down. I mean, just, you know, make sure that a semi-tech literate dope like me understands what the hell you’re saying so we can actually have a conversation.
Paul Triolo: Thanks, Kaiser. I’ll try to do that today. And I want to do make a point that you said that across the AI stack, I’m using the five-layer cake of Jensen Huang these days, and I want to be able to have my cake and eat it too.
Kaiser: Hey, you know, it’s now six layers because we’ve added a layer, thanks to Angela Zhang, I don’t know if you heard my conversation with her — Law is the sixth layer of the AI stack.
Paul: And I also talk about sprinkling that cake in the stack liberally with rare earths.
Kaiser: Samm, Paul, welcome back to Sinica. Welcome to Beijing. And thank you, guys, for doing this in person. It’s so great.
Paul: So happy to be here.
Kaiser: So first, the most important question and the one that everybody is wondering is, you know, how many humanoid robots one can actually watch fold towels before losing the will to live?
Paul: I think it’s the cooking ones that may do that to me, more than the folding the laundry.
Kaiser: Oh, really? Are they good? Did you try anything? I know you’re a very serious vegan, so probably no vegan dishes on offer from the robots.
Paul: No, it was all meat.
Kaiser: They feel no moral anguish when slaughtering animals, I understand.
Samm: I think if we want to judge capabilities, we can look at how the robot dog jumped up higher and did a flip this year versus last year, where it just trotted around.
Kaiser: But I hear it wasn’t just dancing robots this year, huh?
Samm: There were police dogs. There were boxers. And not just at the actual exhibition hall, but in the mall around it where the flocks of people went to escape the heat and humidity, they had a whole match going on between some grandmothers and robots in a boxing ring.
Kaiser: My money’s on the grandmas any day in China, at least. This is a significant upgrade from Rock’em Sock’em Robots, which I used to play when I was a little kid. You remember that, Paul?
Paul: I do. But I focus mostly on, I shouldn’t say, the semiconductor companies and sort of walked past the robots since I did a deep dive into the humanoid robotic sector in April and saw way too many really good humanoid robots.
