DeepSeek’s AI model has shown safety gaps in independent tests, raising questions about its performance under safety assessments.
The tests identified weaknesses in the model’s safeguards, although no further details about the findings were provided.
