DeepSeek Responds to Abnormal Special Character Triggers: Model Hallucination, No Privacy Leak

Recently, some users reported that after inputting special characters in conversations with the DeepSeek model, the model occasionally returns unexpected content, causing some users to worry about "conversation leakage".

DeepSeek's technical team confirmed after investigation that such anomalies are model hallucinations triggered by special characters, with no security or privacy leakage issues involved. The team will enhance the model's ability to recognize and handle special characters through targeted training, fix known problems, and optimize performance in related scenarios.