[IT News] Huawei co-develops DeepSeek-R1-Safe to enhance censorship controls in AI model

[FHISKER]

Huawei, together with Zhejiang University, has created a “safe” version of the DeepSeek-R1 AI model, called DeepSeek-R1-Safe, which includes strong filters for sensitive political content, toxic speech, and topics considered contrary to socialist values, and while in direct tests the model does a near-perfect job of blocking controversial material,

its effectiveness drops significantly in more subtle scenarios like role-playing or encrypted messages, raising questions about how well AI models can adjust to complex social nuances without compromising freedom of expression.

Discover what “safe” AI could look like in the future and what trade-offs it entails!

Source : Reuters Reuters