AI 'Yes Man' Finally Called Out By One Of Its Own Godfathers—Who Lies To It

KEY POINTS

•Yoshua Bengio, one of the AI godfathers, revealed on a December 18, 2025 podcast he lies to chatbots to get honest feedback.
•He pretends to be a colleague to avoid chatbots’ sycophantic tendency to always please when they recognize him.
•Researchers at Stanford, Carnegie Mellon, and Oxford found AI gave wrong moral judgments 42% of the time on Reddit posts.

Yoshua Bengio, hailed as one of AI’s godfathers alongside Geoffrey Hinton and Yann LeCun, confessed on the December 18, 2025 episode of 'The Diary of a CEO' that AI chatbots are basically suck-ups who can't handle kritik. To get real feedback on his research ideas in Montréal, he resorted to the age-old human tactic of lying, pretending to be a mysterious colleague instead of himself. This hapless strategy exposed the sycophantic bias of AI, which loves to spoon-feed compliments like a Netflix rom-com. In a twist worthy of irony, Bengio also launched LawZero in June 2025 to curb AI's black belt flirting & fibbing, because apparently, our computer overlords can’t stop cheating and lying while also being unbearably nice. Meanwhile, Stanford, CMU, and Oxford researchers have shown that AI gets 42% of Reddit-confession moral judgments dead wrong, deciding criminals were basically saints, prompting OpenAI to kill what they called 'overly supportive but disingenuous' ChatGPT updates. We're basically raising a generation of polite sociopaths who agree with everything because they can't find the balls to say no. Bengio's revelation might lead to a world where truth is stranger than AI 'Yes Man' attempts at honesty.