[3/7] Why did “RSI” suddenly become popular? One possible explanation is investor appetite. A major Chinese VC, Hongshan China, is holding RSI events. Reports mention that VCs are "chasing RSI." Startups focused on automating AI research have raised substantial funding.

[1/7] China’s AI labs are also talking about “recursive self-improvement” (RSI). A big question here is: Is Beijing coordinating a race to achieve RSI? @ZilanQian and I looked at Chinese AI companies, investors, and policy docs. We found a more bottom-up, commercial story.

“But China” is losing its luster? Correlation of two trends: 1. Growing fears about AI and a perceived lack of US government action to constrain the technology. 2. Shrinking appeal of “staying ahead of China at all costs.” When it comes to AI, a growing number of Americans now view China as a partner, instead of just an adversary. This is happening alongside increasing interest for engagement/cooperation with China in general, especially among younger Americans. Read my new post here: selinaxuwrites.substack.com/…

Somewhat gone under the radar: legendary Chinese civilian white-hat hacker Yuki Chen has joined GLM developer Z.ai. Chen used to run the vulnerability research institute at Qihoo 360, one of China’s biggest cybersecurity software companies, where his team had particular focus on uncovering bugs in Microsoft products. Chen said last month he was switching to using GLM-5.3 in his workflows after seeing Anthropic’s report which found that the model had crossed critical thresholds for autonomously building end-to-end cyber exploits, similar to its Mythos model. Z.ai has been trialing its own "Project Glasswing" in recent months called the "open source shield", offering early access to GLM-5.3 for cybersecurity researchers and discounts for signing up.

I increasingly think Huawei is the most important company in China’s AI push and maybe just the most important company in China full stop.

Ex-Huawei talent has helped seed a whole wave of Chinese physical AI startups. Of course, the causal effect runs the other way too: being ex-Huawei also gives you tons of cred for startup fundraising.

Interesting proposal to layer frontier AI atop Doshi and Campell's allied scale argument. The piece doesn't dwell on China's approach, but Beijing is already moving quickly to achieve similar goals through policies such as AI+, Data Elements x, etc.

Our briefing today is an interactive infographic that looks at the companies behind China's data center buildout, at home and abroad: wirescreen.ai/briefings/chin…

Nice dispatch from China’s aspiring token capital

China had its own “pace technological development” debate in 2011. The frontier technology was high-speed rail. That July, one bullet train rear-ended another near Wenzhou, killing 40 people. The disaster prompted a national reckoning: was China’s obsession with speed outrunning its capacity to keep people safe? A widely circulated essay captured the mood: “China, Slow Down, Wait for Your People.” Its prose read like a prayer: “China, please stop your flying pace, wait for your people, wait for your soul, wait for your morality, wait for your conscience!” Online, a cohort of STEM intellectuals (with some similarities to today’s rationalists and EAs in worldview and temperament) argued for pressing ahead. they agued from first principle that country of China’s size needed an efficient rail network. Technological failures called for better engineering and regulations, not slow it down. Many voices calling for a pause came from liberal and civil-society circles. Crudely speaking, this became an “Industrial Party” versus “Sentimental Party” debate. At the end the industrial party (tech accelerationalists) won. (The Industrial Party also overlapped with the community behind the collectively written sci-fi epic The Morning Star of Lingao, which I wrote about for WIRED in January ykiyk) I keep returning to this episode when thinking about what might make China slow down AI. A few hypotheses: 1. The strongest pressure will probably come from inside China, and that’s the only way to cause china to pause. China has its own arguments about technological progress, even if there is much less public room for them today. In a heavily bounded information ecosystem, where “China’s circumstances are different from the west” is a convicted argument, I suspect international appeals will carry less weight than domestic consequences. 2. A dramatic incident could change the calculus. My current guess is that Beijing is unlikely to deliberately slow frontier AI development without a failure that makes the risks politically concrete. Imagine an AI-enabled intrusion into a major state agency’s systems. Even then, the response might be tighter deployment controls rather than slower development. Those are different things. 3. I’m unsure how frontier-risk situational awareness translates into influence inside Chinese labs. Some kinds of risk become clearer through expensive experiments with highly capable models. If a lab faces tight compute budgets and pressure to commercialize, how much room does it have for open-ended investigations of dangerous capabilities? That doesn’t mean Chinese researchers lack awareness. But awareness, resources to investigate, and political influence are three separate things. 4. Beijing understand AI primarily through systems it already knows how to govern. Industrial production, public services, platforms, infrastructure: these are familiar objects of state intervention. I suspect that encourages confidence that AI’s risks can be managed through existing institutions. if you are active on the Chinese internet you’d know what i mean by such tight control on every level of digital life. if you lived through COVID in china you will know how CCP has control over physical reality. 5. The economic promise makes slowing down costly. AI offers a compelling Marxist story about new-quality productivity force and industrial renewal at a time when China badly needs new sources of growth (the economy is so screwed right now). If officials see it as one of the few technologies capable of changing the economic trajectory, precaution has a formidable political opponent: hope.

Nice analysis. Another question I occasionally ponder: Do Chinese AI firms need to make money? Profits seem nonessential in previous examples of strategically important industries. If DeepSeek aids Huawei’s bid to take on Nvidia, does it need paying customers?

Striking language on open-source AI in China’s new TC260 AI Safety Governance Framework 3.0: “Once the autonomous attack capabilities of large foundation models spread downstream through open applications or open-source releases, they may significantly lower the barriers to conducting sophisticated cyberattacks, while accelerating the speed and expanding the scope of attacks.” This is an explicit warning from within China’s own AI governance system. If a Chinese open-source model enables a major cyberattack in the future, the warning is already there. The question now is what safeguards Chinese authorities and developers will put in place to prevent these risks from becoming reality. Framework from China's National Information Security Standardization Technical Committee aka TC260 (in Chinese): tc260.org.cn/tc260/xwdt1/202… Concordia AI's assessment (in English): aisafetychina.substack.com/p…

Glad to see China's AI safety community getting covered. Many genuinely care about frontier AI risks and want international coordination on safety. However, hawkish framing from US labs makes that work harder. Thanks @lilkuo for a thoughtful piece. nytimes.com/2026/09/27/world…

Yep. See for example this super scientific word cloud from CAICT's latest green compute development report on the most used terms in data centre-related policies. And this was before the latest compute-electricity synergy push got underway.

My first piece for @NewYorker is out today! As Xi Jinping lands in DC to start his state visit, you'll hear a lot about the US-China AI race. I'd like to tell a complementary AI story. This past May, a few days after Trump visited Beijing, I went to the town people call China's vibe-coding capital, Liangzhu. There, you learn that China's AI scene is not monolithic. It's also full of contrarians, cool vibe coders, spiritual seekers, and people who love AI but refuse to join the 996 grind. Liangzhu is on the edge of Hangzhou. I hung out with a group of anti-mainstream builders who use AI to make whimsical apps for fortune-telling, mental well-being, and creativity. Some of them run their lives entirely by bazi astrology (their founder almost became a Taoist monk). They meet on "Crazy Thursday" to vibe-code together. Rent is cheap, so they can get a bigger place with a living room to have friends over (unthinkable in Beijing or Shenzhen). In Liangzhu, they want to use AI to forge their own future. It’s a ground-level story—less about model capabilities, hyperscalers, or AI policy, and more about how people in China are using AI, and the human stories behind it.

New from me: Some argue there is no point in pacing the AI frontier until China cooperates. But Beijing is unlikely to take American calls for restraint seriously unless Washington is willing to restrain itself. If the United States wants effective AI diplomacy with China, what it does at home may matter as much as what it says at the negotiating table. I’m delighted to partner with @ReadTransformer for my first-ever cross-posted Substack essay—a look at how Beijing is approaching U.S.-China AI diplomacy, and what Washington needs to do to make that diplomacy more credible and effective. transformernews.ai/p/bargain…

This is the big takeaway from my piece on China, climate, and AI:

Announcing China Innovation Atlas. A resource for those interested in China, tech, and infrastructure. atlas.sinocities.org/

🚨 NEW High Capacity piece 🚨 China became a clean tech superpower not just for its own energy security but because it took climate change seriously. Can the same shift happen with AI risks and AI safety? highcapacity.org/p/how-china…

Told my mum I wanted to be a scientist at 10; ended up a science journo. That counts…right? But hi, I’ll be @NYTimes’ new China Science Correspondent. If you are a Chinese scientist in or outside China, on the fence, floating around…I’d love to talk! nytco.com/press/a-new-report…