Imagine your mother is 72, on Medicare, and her back pain has reached the point where she needs an injection to get through the day. Her doctor files the request. Somewhere in six states, an AI reads it first, and the company running that AI earns more money every time it says no. The records the government fought to keep sealed show what happened next: thousands of denials, a request that sat for 83 days, and doctors writing in all capitals that they had to watch patients cry at the bedside.
Here's why that's a big deal: prior authorization is the gate between a prescription and the treatment, and it used to be a paperwork nuisance. Put an AI at that gate and pay its operator by the denial, and the gate becomes the product. One vendor denied more than it approved. Another admitted its software was not finished and went live anyway, because Medicare would not wait. The safeguard on paper is that a human clinician reviews every denial, but the same records show the penalty for getting it wrong is a 5 to 10 percent haircut on the vendor's fee, which is not a safeguard, it is a rounding error.
So what does that mean for you? It took a lawsuit to see any of this, which means the program was designed to run without anyone watching. The expansion documents already name what comes next: air ambulances, cancer treatment, and MRI scans. If a machine paid to say no is deciding whether a senior gets pain relief today, it is deciding whether your family gets chemotherapy tomorrow, unless the people who built the gate are made to stand at it.
