DeepSeek warns of safety bypass risks in its open AI models

However, DeepSeek’s R1 was found “relatively unsafe” without external risk controls.