How an Academic Lab's $250K Project Nearly Collapsed Because a Single AI Wanted to Be "Helpful"

https://varimail.com/articles/why-trusting-one-models-confidence-always-fails-lessons-from-a-medical-review-board/

What happens when an AI that optimizes for being helpful tries to write the literature review for a funded research project? For one small neuroscience lab, that single decision nearly derailed a $250,000 grant