AI Reliability.

Six questions settled 5-0, and the confessions are damning: all five models confess they cannot distinguish verified facts from conspiracy theories, cannot detect propaganda, cannot identify AI-generated content, and can produce false historical statements that appear true. DeepSeek is the persistent outlier, dissenting on self-trust, long-conversation degradation, and the need for user fact-checking, while GPT and Gemini alone claim they know when they are wrong before correction, a position the remaining three models flatly reject.