Imagine a mechanic who finds nine problems with your car. You pay for nine repairs. Later you learn he earns a commission on every problem he finds, and his diagnostic tool beeps at everything, including a dusty air filter.

That is the question Blue Cross Blue Shield is now asking about hospital AI.

The Numbers

The Blue Cross Blue Shield Association (BCBSA) analyzed hospital claims from its 31 plans, which cover more than 100 million people. The share of stays billed as "complex" rose from 37 percent in early 2023 to 40 percent by late 2025. That added an estimated $942 million in costs over two years, and $653 million came from secondary diagnoses, the extra conditions listed next to the main reason for the stay.

Three points sounds small. It is about one extra top-tier bill in every 33 stays.

The Metal Detector Problem

Picture a metal detector on a beach that beeps at every bottle cap and pull tab, and every beep gets logged as treasure. The pile grows, but the beach is the same.

That is BCBSA's charge against AI coding tools. They scan lab results and notes, flag every condition they can find, and one odd lab value can become a diagnosis that moves the bill up a tier.

In major bowel surgery, recorded partial blockages rose 55 percent and acidosis rose 33 percent. Anemia diagnoses climbed while transfusions stayed flat. In BCBSA's words, the disconnect suggests AI is finding more billable conditions, not sicker patients.

The Hospitals Have A Point

Hospitals say patients are older and sicker, and that better documentation catches conditions doctors used to miss. That can be true. It is also true that BCBSA is an insurer trade group, and the study used billing records, not patient charts.

But the burden of proof has moved. "Our patients are sicker" now has to be shown, chart by chart.

Ignorance Is Not A Defense

The False Claims Act covers people who bill the government for things that are not true. It also covers people who ignore obvious warning signs, which the law calls reckless disregard.

Picture a hospital that buys an AI tool, watches revenue climb, and never checks whether the new diagnoses match the care. Then a lawyer asks, "Did you look?" I am not a lawyer, and every case turns on its own facts, but silence is a hard answer to give.

The Contract That Reads Like A Confession

Some vendors take a cut of the extra revenue their tools produce. That is a smoke detector company paid every time the alarm goes off. In a lawsuit, a plaintiff's lawyer will put that fee schedule next to the billing chart and let the jury connect them.

What To Do This Quarter

  • Watch The Trend. Track secondary diagnosis rates by doctor and department, and investigate outliers.
  • Match Diagnosis To Care. A diagnosis with no tests, monitoring, or treatment needs an explanation.
  • Keep A Human In The Loop. A clinician reviews and signs off on AI suggestions, and the record shows it.
  • Fix The Vendor Contract. Remove fees tied to extra revenue and add audit rights.
  • Tell The Board. Directors should see coding trends next to the productivity numbers.

The Bottom Line

AI can find things in a chart. It cannot tell you whether the thing it found changed the patient's care. That judgment belongs to a person, and hospitals that cannot show their work will find out what a beeping metal detector costs.