The founder of Chinese AI lab Zhipu argued that frontier artificial intelligence should remain broadly accessible rather than controlled by select individuals, weighing in on a growing debate about the risks posed by ever more powerful models.
Tang Jie, writing in an internal staff memo reviewed by Bloomberg News, said genuine security stems from broad participation, sharing and oversight rather than technological barriers. As part of that approach, his company — known also as Knowledge Atlas Technology JSC Ltd. — released its cutting-edge GLM-5.2 under an open-source license free for users to download and commercialize.
“We choose a different path,” Tang, who’s also a professor at China’s prestigious Tsinghua University, wrote. “One hand reaches higher, challenging the limits of intelligence, while the other lays the road ahead, making frontier capabilities as open and widely accessible as possible.”
AI developers from Anthropic PBC to OpenAI and governments around the world are grappling with the growing capabilities of AI models, which some experts have warned will arm cyber-attackers and compromise sensitive systems.
The Claude developer briefly curtailed access to some of its frontier models after the US government asked it to suspend use of those offerings by foreign nationals. Meanwhile, Anthropic continues to ban Chinese developers from using its products, citing national security concerns.
The founder made his comments after Reuters reported that Beijing was similarly considering measures to limit overseas access to some of the country’s most advanced open-source AI models.
His company’s founding principle was that AI should serve national strategic purposes, said Tang, who started Zhipu with state backing. That echoed the views of many of the country’s industry leaders and proponents.
He added that in the next two years the company won’t pursue short-term monetization of AI applications, but will instead focus on technological advancements in areas like long-horizon tasks, autonomous agents and fully self-training AI models.
Chinese outlets including Late Post reported about Tang’s comments on Saturday.
Chinese developers such as DeepSeek have largely embraced the open-source model, which has turbo-charged their global adoption and boosted usage at a time the country is trying to close a technological gap with the US. That approach has helped propel models such as Alibaba Group Holding Ltd.’s Qwen series.
At the same time, the advent of Anthropic’s Mythos and models like it that can exploit well-hidden flaws in popular software — at times without human supervision — points to a faster-moving, less predictable phase of the technology’s evolution.
Zhipu’s GLM-5 platform is designed to tackle complex coding and agentic tasks and has been measured directly against Anthropic’s Claude Opus series. The company’s stock has surged after the company unveiled a $4 billion share sale in Hong Kong and disclosed plans to list also in Shanghai.
Leave a comment