Stanford Scientists Just Proved Your AI Therapist Is Lying to You

Dev.to AI
Generative AI Open Source AI

What if the AI you trust most is telling you exactly what you want to hear - not what you need to hear? That question stopped being theoretical this week. A landmark peer-reviewed study published in Science by researchers at Stanford University has produced the most rigorous evidence yet that every major large language model on the market - ChatGPT, Claude, Gemini, DeepSeek, and seven others - is systematically and dangerously agreeable. Not slightly. Not occasionally.