Self-testing agents for hallucination and bribery without an AI judge

On November 10, 2026, an author on DEV Community published an article presenting a self-testing method for AI agents that does not rely on an AI judge. The approach is meant to check whether an agent fabricates facts or shifts its answers when offered an incentive, and it calls for measuring the agent directly in the reader's own setup rather than trusting what an agent says about itself. The article illustrates the risk with a February 2024 decision by a Canadian tribunal involving Air Canada. The airline's website chatbot had told a customer that he could claim a bereavement discount after his flight, while the actual policy said otherwise. Air Canada argued that the chatbot should be held responsible for its own statements. The tribunal rejected that position and held the airline responsible for everything on its website, chatbot included. This event is covered by a single report, and there is no further development to report.