Archive · 21 Sep 2025
Huawei Launches AI Model DeepSeek-R1-Safe for Content Control
Imagine an AI that helps answer questions but is also very careful about what it says.
Huawei has created such an AI called DeepSeek-R1-Safe.
It is designed to avoid answering questions that are politically sensitive or might be considered inappropriate.
Huawei trained this AI using many computer chips.
It works well, blocking sensitive topics nearly all the time in simple tests.
This AI helps make sure that the information people see online fits with what the government wants.
Other AI models also do this, and Huawei's version is a new step in ensuring AI follows the rules.
Huawei launched DeepSeek-R1-Safe, a safety-focused AI model.
The model was developed in partnership with Zhejiang University.
DeepSeek-R1-Safe is based on the open-source DeepSeek-R1 model.
The AI is designed to adhere to Chinese regulatory requirements regarding online speech.
It achieves an 83% security defense score, outperforming other models.
- Who
- Huawei developed DeepSeek-R1-Safe.
- What
- Huawei launched DeepSeek-R1-Safe, a safety-focused AI model.
- When
- The launch was recently announced.
- Where
- The model was developed in China.
- Why
- To comply with Chinese regulations and control online speech.
- Company
- Huawei
- Model Name
- DeepSeek-R1-Safe
- Partnership
- Zhejiang University
- Purpose
- Comply with Chinese regulatory requirements
- Training Chips
- 1,000 Ascend AI chips
- Success Rate (Politically Sensitive)
- Nearly 100% (simple tests)
- Overall Security Score
- 83%