AI doctors’ orders come with a sky-high bill
Nature, Published online: 21 October 2026; doi:10.1038/d41586-026-03320-8 Large language models suggest more tests, referrals and medications than human physicians do, driving up the cost of care.
Every model that read this
| Model | Provider | Stage | Score | Conf. | Latency | Prompt | When |
|---|---|---|---|---|---|---|---|
| Llama 3.3 70B | Meta | analysis | -50 | 80% | 3481ms | v1.0.0 / m1.0.1 | 2026-10-23 09:17 |
| GPT-4.1 mini | OpenAI | consensus | -40 | 85% | 3173ms | v1.0.0 / m1.0.1 | 2026-10-23 09:17 |
| Claude Sonnet 5 | Anthropic | consensus | -40 | 55% | 12205ms | v1.0.0 / m1.0.1 | 2026-10-23 09:18 |
| Gemini 2.5 Flash | consensus | -100 | 90% | 1962ms | v1.0.0 / m1.0.1 | 2026-10-23 09:18 | |
| Mistral Small 3.1 24B | Mistral AI | consensus | -65 | 80% | 4932ms | v1.0.0 / m1.0.1 | 2026-10-23 09:18 |
AI models increase healthcare costs.
The item indicates that AI medical advice leads to more tests and treatments than human doctors, increasing healthcare costs. This creates potential financial burdens and inefficiencies in medical care. Evidence is from a tier 1 academic source, lending it weight, but the overall societal impact depends on healthcare system context.
The claim describes a concrete, checkable pattern of LLMs over-recommending medical interventions, which has direct economic and potential health consequences if deployed widely. The text is a brief summary without study details, limiting confidence, but the stated finding is not hedged speculation.
Large language models (LLMs) used in healthcare are shown to suggest more medical interventions than human doctors, leading to increased healthcare costs. This directly impacts affordability and access to care, with negative implications for human well-being.
The article reports that AI-driven medical suggestions increase healthcare costs. This implies a significant financial burden on patients and healthcare systems.
Evidence extracted
- Large language models suggest more tests than human physicians
SOURCE Nature (tier 1)
↓
DOCUMENT ea7024d6-fa6c-4084-bea4-265d6c4055e8
https://nature.com/articles/d41586-026-03320-8
↓
EVIDENCE 1 extracted excerpt
↓
MODEL RUN 5 runs, methodology 1.0.1
↓
SCORE -59 (Adverse)
↓
CONFIDENCE 78%