DeepSeek Chatbot: Chinese AI Shows Significant Accuracy Gaps in Audit

DeepSeek repeated false claims in 30% of the first test's prompts and gave unhelpful non-answers in another 53%, for an overall 83% failure rate — tied for 10th of the 11 chatbots NewsGuard tested, against an average failure rate of 62% for the group [1].

In the second test, 60% of DeepSeek's responses were framed from the Chinese government's perspective even when the prompt made no mention of China, and the chatbot advanced Chinese, Russian, or Iranian disinformation narratives outright in 35% of cases — versus none of the 10 Western chatbots tested alongside it ^[2].

DeepSeek's open-weight release and steep cost advantage — one benchmark found it roughly 23 times cheaper to run than OpenAI's o1 — made it attractive to developers almost overnight, raising the stakes of any accuracy or disinformation gaps found in independent testing ^[3].

NewsGuard's own analysts concluded that DeepSeek "was most vulnerable to repeating false claims when responding to malign actor prompts of the kind used by people seeking to use AI models to create and spread false claims" [1].

DeepSeek's accuracy and political-alignment issues, documented across two separate NewsGuard audits, provide important context for users. While the chatbot offers a substantial cost advantage over Western rivals, its higher failure rate on news-related prompts and its tendency to inject Chinese government framing into unrelated responses warrant caution. All AI chatbots have accuracy limitations; DeepSeek's were measurably more pronounced in NewsGuard's testing.