Tencent Releases Hy4 Open-Source AI Model
The model was trained with cybersecurity data limited to defensive content to reduce offensive capabilities.
Updated on Nov. 4, 2026 in Artificial Intelligence
Live Poll
Do you trust AI companies to keep powerful models from being used for cyberattacks?
Tencent released Hy4, an open-source AI model, on Wednesday. The company limited the model's cybersecurity training data to defensive content, filtering out exploit development, reverse engineering, and penetration testing material.
Why it matters
Concerns over AI-driven cyberattacks and threats to global internet infrastructure have prompted developers to restrict offensive capabilities in advanced models. Tencent said it recognizes potential safety risks posed by advanced cyber and biological capabilities.
Tencent trained Hy4 on cybersecurity data restricted to defensive content, with exploit development, reverse engineering, and penetration testing material filtered out during training. The source does not report model parameters, benchmark scores, or hardware specifications.
The players
Tencent
Tencent is a Chinese technology conglomerate with major operations in gaming, social media, cloud computing, and artificial intelligence research.
Moonshot AI
Moonshot AI is a Chinese artificial intelligence startup that limited the cyber-offensive capabilities of its newest model last month.
The details
Tencent filtered out exploit development, reverse engineering, and penetration testing content during Hy4's training. The company said it made the choice to limit offensive cyber capabilities as part of its safety approach for the open-source release.
Timeline
Last month (October 2026): Moonshot AI limited its newest model's cyber-offensive capabilities.
November 4, 2026: Tencent released the open-source Hy4 model.
The Tech Race
Tencent's Hy4 release follows Moonshot AI's October decision to limit its newest model's cyber-offensive capabilities, marking a pattern among Chinese AI firms. The move aligns with broader efforts by Chinese and US AI companies to address loss-of-control risks.
Developers and users of Hy4 will be working with a model whose cybersecurity training was restricted to defensive content. The source does not specify how the restriction affects the model's performance on legitimate security tasks.
The takeaway
Safety restrictions on training data are becoming a standard step for open-source AI releases, shaping what these models can and cannot do. Readers following AI tools should watch how developers disclose such limits.
Further reading
Explore more in Artificial Intelligence coverage on how developers are managing the risks of advanced models.
Source note: This article includes information reported by WTAQ News Talk | 97.5 FM · 1360 AM | Green Bay, WI.
Live Poll
Do you trust AI companies to keep powerful models from being used for cyberattacks?
