Link: https://www.straitstimes.com/world/europe/z-ai-founders-worked-on-research-funded-by-china-institute-that-uk-tied-to-spies

Archived (removepaywall.com)

Four founders of one of China’s most promising artificial intelligence companies, Z. ai [also known in China as Zhipu AI], worked on research in a joint project with a technology institute that the British government accused of having “very strong ties” to China’s main spy agency, according to publicly available research papers from China and an “espionage alert” from Britain.

The nearly four-year AI research project, which overlapped with the company’s start in 2019, was co-funded by Beijing-based China General Technology Research Institute (CGTRI) and the National Natural Science Foundation of China, according to a previously unreported 78-page report that the project’s leaders compiled to detail their findings.

A CGTRI engineer and one of Z ai’s founders were among the project’s leaders, and three other Z. ai founders contributed to the research, according to the report.

[...]

An additional report says Z. AI co-founder Li Juanzi, a professor at Tsinghua University, led the research alongside CGTRI engineer Cao Weibing and a Southeast University professor. Three additional Z. AI founders contributed to the work:

• Tang Jie, listed as co-author on project papers
• Xu Bin, named as co-inventor on related patents
• Zhang Peng (current Z. AI CEO), holder of software copyrights from the project

[...]

Separate analysis by national security consulting firm Garnaut Global revealed troubling patterns in Z. AI's handling of software vulnerabilities discovered by its AI models: According to the firm's report, Z. AI established a public ledger on August 14, 2026 documenting 2,436 vulnerabilities found in software from Western technology companies, but disclosed only 5% to affected companies before routing data to Chinese government.

Rather than following standard responsible disclosure practices, Z. AI disclosed only 5% of these vulnerabilities to affected vendors. Ten days after posting the ledger, the company replaced it with a notice stating the Chinese government would manage vulnerabilities through databases including one operated by the Ministry of State Security.

James Mulvenon, who leads intelligence research at national security contractor Pamir Consulting, characterized the situation as giving China's peak cyberespionage agency access to exploitable weaknesses in critical Western software systems.

[...]

Like western frontier model companies!?

No. Models like Mistral and Soofi (an open source AI from Europe) as well as many others have no known links to spy agencies or military.

https://mistral.ai/customers/french-ministry-of-defense/?hl=en-US

Isnt there a law in China that disclosures must be sent to their government first before being sent to the software vendors? So they would be breaking the law if they didnt do that?

Also:

Zhipu AI (Z.ai)

We build the ChatGLM family of LLMs, develop LLMs as Agents, and release related LLM training & inference techniques:

GLM-5.3 & GLM-5.2 & GLM-5.1 & GLM-5, GLM-4.7 & GLM-4.6 & GLM-4.5, GLM-V, GLM-OCR, GLM-ASR, GLM-TTS, GLM-Image, CodeGeeX, CogVLM (VisualGLM), WebGLM, GLM-130B, CogView, CogVideo && CogVideoX. CogAgent, AutoWebGLM, AgentTuning, APAR.

We also work on LLM evaluations: LVBench, MotionBench, ComplexFuncBench, AgentBench, AlignBench, LongBench, NaturalCodeBench.

We started with social networks and graphs, and always love them: GraphMAE, GPT-GNN, GCC, SelfKG, CogDL. AMiner.