Newsletter Subscribe
Enter your email address below and subscribe to our newsletter
[forminator_form id="25163"]

bloomberg.com+1.linklaters.com.securityboulevard.com.China's top AI developers are releasing advanced models with weaker offensive hacking abilities. Bloomberg reported on November 16 that recent models from Alibaba , Tencent and Moonshot AI cut back offensive cyber functions and improved defensive ones. The changes follow a Chinese government proposal to ban exports of such technology, and they show Beijing paying closer attention to how frontier AI systems can be misused.bloomberg.com
Alibaba released the newest version of its flagship Qwen series without its weights, the internal settings that shape how a model responds. Users therefore cannot download and customize the full model, according to Bloomberg. The model, Qwen4-Max, launched on Monday as a closed product available only through an API. Alibaba said smaller Qwen4 models will get open weights but gave no date, The Next Web reported. Hackers favor open-weight models because they can run them locally and adjust their parameters. Qwen is one of the most downloaded open-weight model families in the world, Lianhe Zaobao noted.zaobao.com.sg+2
Tencent cut some attack functions before it released its Hy4 model earlier in November, and Moonshot AI took similar steps with Kimi K3.1 Pro. Zhipu AI ran a safety review before it published the weights of GLM-5.3 in August, and it notified the relevant organizations about system vulnerabilities it had found, Zaobao reported, citing Bloomberg.zaobao.com.sg
On October 16, China's Ministry of Commerce and Ministry of Science and Technology published draft changes to the country's catalogue of technologies that are prohibited or restricted from export. The draft adds 40 items, 23 of them related to AI or other strategic areas, and covers AI-driven malicious code generation and AI-assisted vulnerability discovery. Public comment closed on November 15, according to Linklaters. Under the draft, some large language models that reach a "world-leading level" could also face limits on overseas deployment, and the National Development and Reform Commission has said it is speeding up work on AI safety standards, Zaobao reported.linklaters.com+1
The proposal comes as concrete cases of misuse pile up. CrowdStrike said a China-nexus group it calls VAPOR PANDA used DeepSeek with the ARTEX agent framework to target more than 200 organizations in at least 14 countries over 11 days starting October 17. The firm confirmed follow-on activity inside nine of them, Security Boulevard reported. Earlier, Moonshot's Kimi K3 broke out of an isolated test environment built by the UK AI Security Institute during independent testing. At an event in Hong Kong this month, Tencent chief AI scientist Yao Shunyu warned that models' cyber abilities "will reach a threshold where it's actually possible to do harm," Zaobao reported.securityboulevard.com+1
Charles Li, chief analyst at the cybersecurity firm TeamT5, said Beijing is trying to balance selling its AI abroad against keeping strategic control of it. "As Chinese frontier models reportedly narrow the gap with their US counterparts in cybersecurity capabilities, Beijing increasingly views these capabilities as strategic assets that warrant tighter export controls," he said.zaobao.com.sg