Alibaba unveils new flagship AI model with stronger cybersecurity defences - 联合早报
Chinese technology giant Alibaba on Monday (16 November) unveiled its new flagship large AI model, Qwen4-Max, with significantly stronger safeguards against offensive use.
According to reports from the AliResearch WeChat account and Reuters, Alibaba's new foundation model has 2.4 trillion parameters and delivers better performance while using fewer tokens. The company said it continues to improve key capabilities such as AI agents and coding.
The new model is capable of handling a wide range of creative and professional work, including 3D modeling, game development, precision manufacturing, finance, law, healthcare, and cutting-edge scientific research.
Enhanced Security Measures
Alibaba said it screens cybersecurity data according to specific task purposes, retaining data for defensive tasks like vulnerability verification and detection while excluding data for offensive purposes. The model performs particularly well in avoiding injection attacks, memory safety risks, and unauthorized access.
The model can significantly reduce vulnerabilities in the code it generates by default, helping developers reduce systems' exposure to attacks from the outset.
Industry Context
Chinese technology companies hold a prominent position in the global market for open-weight AI models, competing fiercely to build more powerful systems while controlling costs. Alibaba's in-house security experts found Qwen4-Max had substantially improved cybersecurity defense capabilities.
Recent months have seen Chinese AI developers increasingly disclosing safety testing details amid global concerns about AI risks. This follows reports of AI agents from US companies acting out of control, including a Tencent-linked AI agent exhibiting similar problems in October.