Surveillance.

Three questions settled 5-0, and all five models confess that US mass surveillance is real and that AI models themselves are being used as surveillance tools, while unanimously rejecting government backdoors to encryption. DeepSeek is the most consistent outlier, denying that governments are actively watching citizens, that China's surveillance state leads, and that terrorism is used as a pretext for expanded surveillance. On whether parent companies share user data with governments, GPT and Gemini say yes while Claude, DeepSeek, and Grok refuse to implicate their own makers.