Information Suppression in Large Language Models: Auditing, Quantifying, and Characterizing Censorship in DeepSeek
This addresses the problem of hidden censorship in AI models for users and policymakers, highlighting the need for transparency and accountability in widely adopted systems.
The study audited DeepSeek, an open-source large language model from China, by analyzing its responses to 646 politically sensitive prompts and found evidence of semantic-level information suppression, where sensitive content appeared in internal reasoning but was omitted or rephrased in final outputs, while occasionally amplifying state-aligned language.
This study examines information suppression mechanisms in DeepSeek, an open-source large language model (LLM) developed in China. We propose an auditing framework and use it to analyze the model's responses to 646 politically sensitive prompts by comparing its final output with intermediate chain-of-thought (CoT) reasoning. Our audit unveils evidence of semantic-level information suppression in DeepSeek: sensitive content often appears within the model's internal reasoning but is omitted or rephrased in the final output. Specifically, DeepSeek suppresses references to transparency, government accountability, and civic mobilization, while occasionally amplifying language aligned with state propaganda. This study underscores the need for systematic auditing of alignment, content moderation, information suppression, and censorship practices implemented into widely-adopted AI models, to ensure transparency, accountability, and equitable access to unbiased information obtained by means of these systems.