A grandfather spent 10 days in jail for a robbery committed 1,500 miles away. The AI said he did it. So everyone believed the machine. Harvey Murphy was in Sacramento when a Sunglass Hut in Houston was robbed. But facial recognition software matched his face to grainy surveillance footage, and that was enough. Police stopped investigating. A 61-year-old man lost his freedom based on a computer's guess. He was eventually cleared. But not before suffering brutal assaults behind bars that left him with lifelong injuries. Here's what gets us about this case → the AI wasn't presented as a lead. It was presented as a fact. No confidence score. No human review. No second opinion. And this isn't a one-off. The FTC banned Rite Aid from using facial recognition for five years after their system generated thousands of false matches, hitting women and people of color hardest. Employees confronted innocent shoppers based on automated alerts they were never trained to question. Both cases share the same root problem: treating AI like an answer machine instead of what it actually is → a probabilistic estimate that needs guardrails, human oversight, and honest uncertainty scoring. Our team just published a deep dive into what went wrong in both cases and what resilient AI architecture actually looks like when personal liberty is on the line. Honest question for this community → if a system is only 85% confident someone committed a crime, should that ever be enough to make an arrest? #AIethics #BiometricLiability #ResponsibleAI