DeepSeek模型在中文语境下的安全性评估（英文版）

Abstract. Recently, the DeepSeek series of models, leveraging their ex-ceptional reasoning capabilities and open-source strategy, is reshapingthe global AI landscape. Despite these advantages, they exhibit signif-icant safety deficiencies. Research conducted by Robust Intelligence, asubsidiary of Cisco, in collaboration with the University of Pennsylva-nia, revealed that DeepSeek-R1 has a 100% attack success rate whenprocessing harmful prompts. Additionally, multiple safety companies andresea