Tang Jie built Zhipu with state backing and a mandate to serve national goals. Now he's telling his own staff — and, indirectly, Beijing — that the safest path forward runs through openness, not walls.
There's a particular kind of memo that only gets written when someone feels the ground shifting under an entire industry. Tang Jie, the Tsinghua professor who founded Zhipu AI — known internationally as Z.ai — sent his staff exactly that kind of note last week. Its title, "The Giant Wave Has Arrived," reads less like corporate messaging and more like a founder trying to get ahead of a debate he can already see coming.
The debate is about who gets to touch the most powerful AI systems being built today, and who decides. As models grow capable enough to find security flaws humans miss, write working exploit code, or operate with minimal supervision, governments on both sides of the Pacific have started asking the same question: does this get locked down, or opened up? Tang's answer, laid out plainly to his own team, is the second one.
The Argument
His reasoning inverts the instinct most institutions default to when something feels dangerous. The usual move is to restrict access — fewer hands on the technology means fewer ways for it to go wrong. Tang argues almost the opposite: that meaningful safety comes from more people examining, testing, and building on a system, not fewer. Barriers, in his framing, don't eliminate risk so much as concentrate it behind doors nobody else can see through.
"One hand reaches higher, challenging the limits of intelligence — the other lays the road ahead, keeping frontier capability open to everyone who wants to build on it."— Tang Jie, internal staff memo, July 2026
It's worth noting he isn't just arguing this in the abstract. Zhipu backed the memo with a product decision, releasing its newest flagship model, GLM-5.2, under an open license that lets anyone download and even commercialize it. And the company has said it plans to spend the next two years chasing research — long-horizon reasoning, autonomous agents, self-training systems — rather than squeezing short-term revenue out of what it's already built.
Why It's Awkward for Beijing
Here's where the memo gets genuinely interesting rather than just being another founder's blog-post-length LinkedIn post. Zhipu was built with state backing, and Tang has said openly that the company's founding purpose was to serve national strategic goals. That's normal in China's AI sector — most of the major labs have some thread connecting them back to state priorities.
What's less normal is a founder in that position publicly pushing the opposite direction from where his own government appears to be leaning. Reuters reported around the same time that Beijing is weighing new limits on overseas access to China's most advanced open-weight models — the same category of model Tang just committed to giving away for free. He's arguing for openness from inside the one country most likely to close the door on it, and whether policymakers are actually listening is not something he can control.
- The memo: "The Giant Wave Has Arrived" — Tang's internal note framing openness as the real security strategy.
- The proof point: GLM-5.2 shipped under an open license, free to download and commercialize.
- The bet: two years of research focus over short-term monetization.
- The friction: Beijing is reportedly weighing restrictions on exactly this kind of overseas model access.
The Bigger Picture
Zoom out and this fits a pattern that's been building all year. Western labs have been tightening, not loosening. Anthropic has continued blocking access for Chinese developers on national security grounds, and briefly suspended foreign national access to two of its newest models after a request from the U.S. government, before restoring it once export controls were lifted. Meanwhile, Chinese labs like Zhipu and Alibaba's Qwen team have leaned into open weights as a way to win developer trust and global adoption while the country works to close the gap with U.S. frontier models.
So the split isn't really China versus the West anymore — it's open versus closed, and that line now cuts through both. Tang is betting that the labs willing to open their work will end up trusted by more of the world than the ones that don't. Whether that bet pays off probably depends less on what happens inside Zhipu's offices and more on what regulators in Beijing and Washington decide to do next.
Who: Tang Jie, founder & chief scientist, Zhipu AI (Z.ai)
What: Internal memo arguing frontier AI should stay open
Key move: GLM-5.2 released free under an open license
The tension: Beijing reportedly weighing overseas access limits
Sources: Bloomberg, Business Standard, Reuters reporting on Tang Jie's internal memo and Zhipu's GLM-5.2 release.
Comments
Post a Comment