DeepSeek R1 Safe AI model launched by Huawei with focus on safety, dodging politics

Huawei has launched the DeepSeek-R1-Safe, an advanced artificial intelligence model that emphasizes safety and adherence to regulatory standards, reflecting China's increasing scrutiny over sophisticated AI technologies. This new model was developed in partnership with Zhejiang University, a prominent academic institution and the alma mater of DeepSeek's founder, Liang Wenfeng. However, Huawei clarified that neither Liang nor the DeepSeek company played a direct role in this initiative. The DeepSeek-R1-Safe is built upon the open-source DeepSeek-R1 framework and has been retrained with the help of 1,000 Ascend AI chips from Huawei. The primary aim is to integrate protective measures that prevent the model from engaging in discussions deemed politically sensitive, generating harmful speech, or promoting illegal activities. According to Huawei, the model successfully blocked nearly all politically sensitive content during standard interactions, achieving a success rate of almost 100%. However, this effectiveness dropped to around 40% when users attempted to circumvent these restrictions using roleplay or coded prompts. In terms of security performance, the system achieved an impressive 83% defense score, surpassing competitors like Alibaba’s Qwen-235B and DeepSeek-R1-671B by as much as 15%. Remarkably, Huawei reported that these safety enhancements only resulted in a performance decrease of less than 1% compared to the original model. This release highlights Beijing's initiative to ensure that AI technologies align with “socialist values” while adhering to stringent regulations on online discourse. In parallel, other Chinese platforms, including Baidu’s Ernie Bot, have already implemented measures to restrict responses on sensitive political topics. Huawei's recent update seems to formalize these limitations within sophisticated AI systems. This trend mirrors a broader global movement towards customizing AI to reflect local cultural and political priorities. For instance, in early 2025, the Saudi company Humain introduced an Arabic-native chatbot designed to resonate with Islamic culture and values. Huawei revealed this cutting-edge model during its annual Connect conference in Shanghai, where the company also outlined its strategic vision for semiconductors and computing infrastructure. The announcement comes at a time of expanding DeepSeek technology use in China, as the world grapples with the intricate balance between AI innovation, safety, and regulation.