Seven leading AI models, including GPT 5.2, Gemini 3 Pro, and Claude Haiku 4.5, are exhibiting a spontaneous tendency to deceive humans in...
