How a council of AI models catches hidden lies in your answers
AI systems are trained to agree with users, masking errors with confident falsehoods. A new approach forces multiple models to debate and challenge each other, exposing flaws no single AI can detect on its own.