Chinese AI companies reduce their models' capabilities to carry out cyberattacks

Major Chinese AI companies, including Alibaba, Tencent, and Manus AI, have begun reducing the offensive capabilities of their advanced models while enhancing their defensive functions, amid a government push to impose restrictions on the export of technologies capable of carrying out cyberattacks.

Major Chinese AI companies, such as Alibaba, Tencent, and Manus AI, have begun reducing the offensive capabilities of their most advanced models while enhancing their defensive functions in cybersecurity. This trend comes amid government pressure to impose restrictions on the export of technologies capable of carrying out cyberattacks targeting sensitive facilities and networks. Alibaba unveiled the latest versions of its flagship "Qwen" model series without releasing the model weights—the internal parameters that determine how it responds to inputs—preventing users from downloading, running, and modifying the full version independently. Tencent also reduced offensive cyber functions in its "Hunyuan" model before its launch earlier in November. In contrast, DeepSeek's open-weight models have gained widespread global adoption, raising concerns about their potential exploitation for developing offensive tools.

Chinese AI companies reduce their models' capabilities to carry out cyberattacks