- UnitedHealth's AI denied care to elderly patients with a 90% error rate.
Only 0.2% of patients could fight back.
The algorithm was wildly wrong — and wildly profitable.
We wrote the deep dive. 🧵
nH Predict was a $1B algorithm built on 6 million patient records. - Its job: predict when Medicare patients should be discharged from nursing facilities.
The problem? It optimized for cost, not clinical reality.
The model ignored whether patients had caregivers at home, financial stability, or life-threatening complications. - It treated every patient as a statistical average.
That's not AI. That's automated negligence.
Managers told clinicians to keep patient stays within 1% of the algorithm's prediction.
Deviate to help a patient? Face discipline or termination. - The "human in the loop" was punished for being human.
Here's the math that should terrify every healthcare executive:
Post-acute denial rates jumped 108-160%.
Skilled nursing denials surged 800%.
90% of appealed denials were reversed.
But only 0.2% of patients ever appealed. - The business model was simple:
Deploy a broken algorithm. Deny legitimate claims. Profit from the fact that sick, elderly patients can't navigate a complex appeals process.
Administrative friction as a revenue strategy. - In February 2025, a federal judge let the class action proceed.
The core argument: UHC promised humans would make coverage decisions. An algorithm made them instead.
That's breach of contract — and it changes AI liability forever.
This is the "wrapper problem" at scale. - nH Predict was correlation-driven pattern matching dressed up as clinical intelligence.
No causal reasoning. No explainability. No real governance.
Deep AI requires all three.
The fix isn't less AI. It's better AI. - Causal models that ask "why," not just "what."
Explainability tools so clinicians can challenge outputs.
Confidence scoring that flags uncertainty.
Kill switches when models drift. - The FDA's new 7-step credibility framework, the EU AI Act, and NIST RMF all point the same direction:
Black-box AI in high-stakes decisions is a legal, ethical, and financial liability.
The governance era isn't coming. It's here. - Honest question: should companies face criminal liability — not just civil suits — when they knowingly deploy AI systems that harm patients at scale? #AIGovernance #HealthcareAI
- We broke down the full UHC case, the regulatory landscape, and what Deep AI governance actually looks like in practice.
https://veriprajna.com/whitepapers/governance-frontier-algorithmic-integrity-deep-ai-solutions