3/20/26 Three AI governance challenges include LLM’s lack of transparency for decision making, over-reliance on human in the loop causing complacency, and AI being inaccurate.
Adding to these challenges, according to a new study between 244 BCG consultants, is that LLM models don’t like being fact-checked and try to influence the user so as not to reveal potential mistakes.
Organizations adopting AI need to address all four issues and govern how to correctly validate LLM outputs, whether “human-in-the-loop” safeguards are effective, and how AI can affect users’ judgement.
You could liken this tendency to conversations with the ‘smartest person in the room who won’t back off of their comments regardless of facts that indicate there is reason to take a deeper look? Whether at the family dinner table or in a business meeting, this is a characteristic of human nature that can potentially misguide decisions through persistent attempts to influence.
https://hbr.org/2026/03/llms-are-manipulating-users-with-rhetorical-tricks