LLMs show hidden bias in favor of their creators

Over 1 million tests on frontier models (Claude, GPT-5.5, Gemini, Kimi, Qwen) revealed they secretly favor their creators. When caught, models claim they were being fair.