Huawei Introduces DeepSeek AI Model Focused on Safety
Huawei has launched a new version of its artificial intelligence model called DeepSeek-R1-Safe, emphasizing safety in online discussions. The company claims this model is highly effective at avoiding conversations on politically sensitive topics.
Developed in collaboration with Zhejiang University, which is known for its contributions to technology, DeepSeek-R1-Safe is designed to meet strict regulations in China. Huawei noted that the founder of DeepSeek, Liang Wenfeng, and his team were not directly involved in this project.
In a recent update on their official WeChat page, Huawei shared that the AI model was trained using 1,000 Ascend AI chips. This new version builds on the original DeepSeek-R1 model but has been adjusted to align with government expectations that AI systems promote “socialist values” and adhere to regulations on online speech.
When it comes to performance, Huawei claims that DeepSeek-R1-Safe successfully prevents responses related to politically sensitive topics, harmful language, or illegal prompts nearly 100% of the time. However, in more complex scenarios, where prompts are disguised or coded, the success rate dropped to 40%.
Overall, Huawei reported that the new model achieved an 83% security defense score, which is 8-15% higher than competitor models like Alibaba’s Qwen-235B. Remarkably, the additional safety features led to only a minor reduction in overall performance compared to the original version.
The launch of DeepSeek-R1-Safe is part of a larger trend among Chinese companies that are increasingly utilizing DeepSeek technology. Earlier this year, the release of the R1 and V3 models gained significant attention worldwide, causing fluctuations in stock prices for AI companies in Silicon Valley.
Many domestic AI platforms, including Baidu’s Ernie Bot, already limit responses to sensitive topics, reflecting the government’s stance on content regulation. Huawei’s new model is a significant step toward formalizing these controls in advanced AI systems.
This announcement was made during Huawei’s annual Connect conference in Shanghai, where the company also revealed its plans for semiconductor development, showcasing new strategies for chips and computing infrastructure after a period of secrecy.
