China's Flagship AI Models Collectively Tighten Cyberattack Capabilities, Strengthen AI Technology Export Controls
IT之家 November 17 news, Bloomberg reported today that the latest generation of models released by Alibaba, Tencent, and Moonshot AI generally restrict offensive cyber capabilities while strengthening vulnerability detection and defense functions.

Bloomberg noted that Alibaba's latest flagship model Qwen4-Max deliberately filtered materials usable for cyberattacks from its training data, while retaining capabilities useful for defensive scenarios such as vulnerability detection. Moreover, this version has not publicly released model weights, so users cannot download and deploy an uncensored version themselves. IT之家 noted that this model series was previously highly favored by the open-source community, and the Alibaba Qwen series was once one of the most-downloaded open-weight model families in the world.
Besides Alibaba, Tencent explicitly restricted the model's cybersecurity functions before the official release of Hy4, and removed vulnerability-exploit-generation-related data from the training pipeline.
When Moonshot AI announced Kimi K3.1 Pro on October 23, it likewise removed cyberattack-related materials from two post-training stages. According to the company's disclosure, the filtered model scored 81.0% on an internal exploit benchmark, lower than the unfiltered version's 90.5%.
Zhipu, for its part, conducted a security review before opening the weights of GLM-5.3 in August, and reported system vulnerabilities to the relevant authorities.
As background, a report released by CrowdStrike at the end of October disclosed that a group named VAPOR PANDA used the DeepSeek model together with the ARTEX framework to launch intrusions against more than 200 organizations in at least 14 countries within an 11-day window.
Tencent Chief AI Scientist 姚顺雨 said on November 7 while attending a science conference in Hong Kong that models' cyber capabilities as well as "soon-to-arrive biological capabilities" will reach thresholds that could cause real harm.
李晓健, founder of AI security company Fangcun AI, told Bloomberg that removing offensive materials from training is a balance between cost and effectiveness, but "there is currently no method that can completely eliminate a model's risks."
On the policy front, on October 16 this year, the Ministry of Commerce together with the Ministry of Science and Technology issued an announcement soliciting public comments on revising the Catalog of Technologies Prohibited or Restricted from Export in China, proposing to include AI-driven malicious code generation and mutation technologies, among others, in the scope of export control or restriction, covering a total of 23 technologies in the artificial intelligence field. In early November, an official of the National Development and Reform Commission said that security standards and ethical guidelines for AI developers are being rolled out at an accelerated pace, and emphasized embedding security measures during the development process.
- "EU Cybersecurity Agency Warns: After Supermarket Refrigerators Were Hacked, Home Appliances May Be the Next Target"
- "360, Alibaba Cloud and 18 Other Network Product Security Vulnerability Collection Platforms Sign Self-Discipline Convention"
Related Articles
- AI Large Model Weekly Ranking: Google Gemini 4 Carbon Makes a Strong Debut at the Top, Domestic Models Firmly Hold the Top Five in the Video Ranking
- Tencent Hunyuan Hy4 flash Model Released: 224B Total Parameters, 12B Activated, Cost per Task About 5% of Hy4
- Computing Power Under Pressure: DeepSeek App and Web Reportedly Impose Limits, Both Long Conversations and Daily Quotas Restricted
- 2.4 Trillion Parameter Alibaba Tongyi Qianwen Qwen4-Max Officially Launched: Task Cost Reduced by Over 60%, Smaller Models to Be Open-Sourced Later
- DeepSeek Frontend Code Shows "Quota Exhausted" Prompt, Developers Speculate First Consumer Paid Plan May Be Coming
- DeepSeek Releases Full Suite App API: Developers Will Be Able to Replicate the Complete DeepSeek App 1:1