3.72 Million Times AI Broke — And No One Was Watching

In 16 months, Americans filed 3.72 million reports that an AI system failed them — outages, wrong answers, dead agents mid-task. The data comes from Ookla, the company that owns Downdetector, in a June 10, 2026 report bluntly titled "AI Reliability Risk Is Now Business-Critical."

This isn't a chatbot going quiet. Agentic AI now books meetings, moves money, and controls devices — and it broke at a 210% higher rate than basic AI tools. When one piece failed, it cascaded: downstream systems stalled, manufacturing lines stopped, fraud detection went dark. 68% of enterprises now run at least one AI-dependent business process. 41% run three or more. Most companies still treat AI reliability as "an afterthought, bolted onto existing IT operations."

No regulator has acted. Downtime-reporting mandates are only "on the horizon." Meanwhile Ookla projects the numbers could double in 18 months. We didn't build the oversight. We just plugged the AI in and walked away — and the bill is 3.72 million reports and counting.