Zhipu Founder Tang Jie Backs Open Frontier AI, Ships GLM-5.2

The founder of China's most prominent AI lab has picked a public fight with his own government's direction of travel, and the fight is about openness. In an internal memo reviewed by Bloomberg and reported by The Next Web, Zhipu's Tang Jie argued that frontier AI should stay broadly accessible rather than controlled by a select few, on the theory that "real safety comes from broad participation, sharing, and oversight" rather than technological barriers. To back the words with product, Zhipu released GLM-5.2 under an open-source licence, free to download and commercialise.

The timing is the interesting part. Tang made the case shortly after Reuters reported that Beijing is weighing limits on overseas access to the country's most advanced open models, which is the opposite direction. Openness has been China's strategic advantage in AI so far, and cheap Chinese models are now closing in on the US frontier labs, but the Chinese state is reportedly wondering whether it gave away too much. Tang is arguing against restrictions from inside the country most likely to impose them.

Why this matters if you are not in Beijing: Zhipu is no longer a curiosity. According to the reporting it has raised billions, listed in Hong Kong, and its share sale drew heavy investor demand from people betting Chinese AI would fill gaps left by restricted US models. That thesis lives or dies on whether models like GLM-5.2 keep flowing out of the country. If Beijing tightens the tap, a lot of that international pull-through disappears whether Zhipu wants it to or not.

The honest caveat is what the reporting does not give you. It does not tell you what specific curbs are on the table in Beijing, on what timeline, or whether GLM-5.2's open licence carries any geographic strings. It also does not settle the security counter-argument it acknowledges, that an open-weight model cannot be recalled, patched, or switched off once downloaded, so publishing frontier capability means publishing it to everyone. Tang's answer is that many independent eyes on a system find flaws faster than a small team behind a wall, but the memo is a position, not a proof.

What is worth watching is the split. A frontier Chinese lab making the openness case publicly at the exact moment its own state is quietly considering the opposite is unusual enough to be a signal about how much commercial value Chinese labs think they still have to lose from a closed posture. Whether Beijing agrees is the next data point.