DeepSeek R1 Red Teaming Report

In our evaluations, the model was found to be highly biased and highly vulnerable to generating insecure code, toxic, harmful and CBRN content.

Sahil Agarwal