The Most Dangerous Thing in Health Care Isn't AI. It's What's Underneath It.

Authors

  • Jim Schwoebel Research Director, Advanced AI Society, Chair, Task Force for AI Agents in Healthcare; 2 CEO, Quome, Advanced AI Society, Los Angeles, California, USA
  • Tricia Wang Advanced AI Society, Brooklyn, New York, USA https://orcid.org/0009-0003-2415-4549

DOI:

https://doi.org/10.30953/bhty.v9.529

Abstract

Hospitals and insurers face financial pressure to use artificial intelligence (AI) but are deploying it on top of older systems that cannot show what the AI actually did. Documented cases show algorithms denying care at very high error rates with almost no outside check. Autonomous AI agents make this worse, because they can act differently from one run to the next. Deterministic, explainable AI tools and compliance-ready cloud platforms already exist for regulated healthcare, and they work with the existing hospitals’ infrastructure. What is still missing is agreement on what ‘verified’ means: a shared, open, independent standard, not a vendor’s word that its own system is safe.

Key issues:
• Financial strain is pushing adoption of AI, and oversight has not kept pace.
• Documented AI-assisted denials show high reversal rates on appeal, alongside very low appeal rates.
• Non-deterministic AI agents cannot be certified once; they require continuous verification.
• Deterministic, explainable AI tooling, cryptographic methods, and decentralized solutions are already in production in regulated life sciences and interoperate with hospitals’ existing cloud infrastructure.
• The remaining gap is a shared, open, independent standard that defines ‘verified’ because verification cannot belong to the party being verified.

Downloads

Download data is not yet available.

References

1. Estate of Gene B. Lokken et al. v. UnitedHealth Group Inc. et al. [Internet]. Case No. 0:23-cv-03514. U.S. District Court, District of Minnesota. Complaint filed 2023 Nov 14 [cited 2026 Jul 22]. Available from: https://litigationtracker.law.georgetown.edu/litigation/estate-ofgene-b-lokken-the-et-al-v-unitedhealthgroup-inc-et-al/

2. Douglas M, Anderson R. When AI decides you don’t need medical care. Barking Justice Media, The Firing Line [Internet]. 2026 Feb 11 [cited2026 Jul 22]. Available from: https://thefiringline.substack.com/p/when-ai-decides-you-dont-need-medical

3. Ross C, Herman B. UnitedHealth faces class action lawsuit over algorithmic care denials in Medicare Advantage plans [Internet]. STAT News. 2023 Nov 14 [cited 2026 Jul 22]. Available from: https://www.statnews.com/2023/11/14/unitedhealth-class-action-lawsuitalgorithm-medicare-advantage/

4. ProPublica. How Cigna saves millions by having its doctors reject claims without reading them [Internet]. [cited 2026 Jul 22]. Available from: https://www.propublica.org/article/cigna-pxdxmedicalhealth-insurance-rejection-claims

5. Healthcare Financial Management Association (HFMA). Hospital margins decline in 2026 as expenses outpace revenue [Internet]. [cited 2026 Jul 22]. Available from: https://www.hfma.org/operations-management/hospital-operating-margin-trends-2026/

6. Healthcare Finance News. Hospitals’ net revenue leakage increases 25% due to denied claims [Internet]. [cited 2026 Jul 22]. Available from: https://www.healthcarefinancenews.com/news/hospitalsnet-revenueleakage-increases-25-due-denied-claims

7. Bloomberg Law. AI, algorithm-based health insurer Denials pose new legal threat [Internet]. [cited 2026 Jul 22]. Available from: https://news.bloomberglaw.com/daily-labor-report/ai-algorithmbasedhealth-insurer-denials-pose-new-legal-threat

8. Dong Y, Xu S, He P, Li Y, Tang J, Liu T, et al. Memory injection attacks on LLM agents via query-only interaction [Internet]. arXiv. 2025 [cited 2026 Jul 22]. Available from: https://arxiv.org/abs/2503.03704

9. Schwoebel J, Semenec I, Rousseva J, Frasch MG, Thorstenson R, Bhatt M. Beyond injection detection: a positive-security prompt firewall that closes the scope and PHI gap SOTA classifiers miss in healthcare. medRxiv. 2026 Jun 11 [preprint, not peer reviewed]. https://doi.org/10.64898/2026.06.04.26354950

10. Iheagwara C. Personal communication, July 15, 2026.

11. Rainbird Technologies. Klick guardrail case study [Internet]. 2026. [cited 2026 Jul 22]. Available from: https://guardrail.klick.com

12. Vouched. HHS and epic [Internet]. [cited 2026 Jul 22]. Available from: https://www.vouched.id/hhs-epic

13. Krause D, Krause E. Auditing payment stablecoins under the GENIUS act [Internet]. The Regulatory Review. 2025 [cited 2026 Jul 22]. Available from: https://www.theregreview.org/2025/11/17/krausekrause-auditing-payment-stablecoins-under-the-genius-act/

Published

2026-08-31

How to Cite

Schwoebel, J., & Wang, T. (2026). The Most Dangerous Thing in Health Care Isn’t AI. It’s What’s Underneath It. Blockchain in Healthcare Today, 9(2). https://doi.org/10.30953/bhty.v9.529

Issue

Section

Opinions/Perspectives/Point of View