AI briefing: Hyped AI goes to the hospital | Is healthcare AI sustainable? | US & Israel power up AI partnership | more
Just because a healthcare AI product is hyped doesn’t mean its claims are bogus.
The salient question is whether or not it performs as well in clinical settings as it did in a research lab—or in a marketer’s mind.
The challenge of distinguishing contenders from pretenders gets fleshed out in a new literature review from the ARISE network. That’s the multi-institution research collaboration anchored by faculty physicians at Stanford and Harvard with network co-founders from the Universities of Virginia and Minnesota. The report, “The State of Clinical AI (2026),” shows the team scrutinized select studies of clinical AI published in 2025. The papers they reviewed had high impact scores and included outcomes reports for such real-world performance indicators as diagnostic accuracy, bias mitigation, workflow effects and user performance.
Peter Brodeur, MD, Ethan Goh, MD, and colleagues state they designed the project to look “beyond model performance.” That is to say they took into consideration such clinically consequential factors as how systems are evaluated, how clinicians and AI work together, and where patient risks start to appear in hospitals, clinics and other care settings. “Frontier AI systems are already powerful,” they note in an introductory message. “What’s needed now is to safely and effectively translate these tools into real-world care.”
Among the team’s big-picture takeaways:
- Model capability is accelerating, but evidence of real clinical impact remains limited. “Many studies show what models can do in controlled settings,” the ARISE researchers write. “What’s increasingly needed are prospective studies that show measurable effects on patient outcomes and care delivery.”
- Frontier LLM models show very uneven performance. “They perform extremely well on complex reasoning tasks yet break down when uncertainty, missing information or changing context is introduced.”
- Clinicians value automation where it reduces administrative and workflow burden. However, these use cases remain understudied, the authors remark. “Tasks clinicians most want support with are often underrepresented in current benchmarks and evaluations.”
- Patient-facing AI has significant potential to reshape engagement and access, but it raises distinct safety concerns. “Direct interaction with patients requires much stronger guardrails and scalable oversight systems that do not currently exist.”
- Multimodal clinical AI applications are approaching practical usability. “Improvements in base models are enabling applications that integrate unstructured text, images and other clinical data to support prediction and decision-making in real-world settings.”
- FDA clearance is increasing, but near-term clinical adoption will favor narrow, task-specific systems. “AI tools that are tightly scoped to specific domains and contexts are more likely to demonstrate value and be adopted in practice.”
- Healthcare AI’s next phase will not be driven by newer models alone. This point is from coverage of the study by Stanford Medicine’s news operation. In the article, reporter-researcher Rebecca Handler, MSc, says the new ARISE research makes clear that the near-term future of healthcare AI “will depend on whether health systems, researchers and regulators are willing to apply the same standards of evidence to AI that they expect of any other clinical intervention.”
Healthcare AI is potentially wonderful. But is it environmentally sustainable?
It can and should be both. However, if AI-using provider organizations fail to appreciate the sustainability challenges involved, the technology could help heal patients while harming the physical world in which we all live. Researchers issue a call for action in a peer-reviewed article published Jan. 15 in Health Affairs. “Clinicians should learn about sustainable AI practices to ensure responsible use of AI in their practices,” write Manijeh Berenji, MD, MPH, of UC-Irvine and colleagues at Yale, Weill Cornell and the University of Michigan. “Likewise, hospital administrators should advocate for sustainability considerations when developing internal AI models and encourage greater transparency from AI companies they contract with about carbon emissions. While AI holds significant promise for improving healthcare, sustainability must guide its application.” Key excerpts from the piece:
- AI models emit carbon dioxide with every query. The data centers processing healthcare AI prompts use a lot of electricity, which will probably continue to burn fossil fuels for the foreseeable future. “ChatGPT, for instance, consumes as much as 40 watt-hours of electricity when generating a medium-length response of about 1,000 tokens (words or parts of words),” Berenji and co-authors point out. “At roughly one billion queries daily, its energy use each day equals the annual consumption of about 1,450 households.”
- About 65% of U.S. hospitals have used predictive models. Among them, 79% have relied on models supplied by their EHR vendor. Many large hospitals have also developed their own predictive models using proprietary patient data sets, the authors note. “In the present day, it is difficult to precisely quantify AI-associated emissions, as models’ carbon footprints vary widely based on the type of hardware used, model size, number of training runs, energy mix of the data center, data center efficiency and usage volume,” they write. LLM-driven patient communications are a good example of an application that’s small enough per message but massive at scale. “The overall computational and operational impact,” the authors write, “is driven by high message volumes across large patient populations.”
- Hospital administrators should question the sustainability of AI models they purchase. They can do so by inquiring with the vendor about carbon emissions involved in the development, training, implementation and use of the AI technologies at hand. “Because technology companies have a vested interest in their customers’ satisfaction, hospital administrators are uniquely positioned to ask about sustainability considerations and to advocate for policies that encourage the development of sustainable AI,” Berenji and co-authors comment. “Furthermore, scrutinizing the carbon footprints of different AI models allows decision-makers to partner preferentially with companies that incorporate sustainability considerations into model development. Competition would provide pressure for other AI companies to adopt more sustainable practices.”
The U.S. and Israel are newly partnering on AI and related technologies.
Representatives of both countries gathered in Jerusalem Jan. 16 to release a joint statement about the deal. At the modestly publicized event, U.S. Undersecretary of State for Economic Affairs Jacob Helberg noted the agreement tasks the nations to “move beyond admiration and into production.” Specifically, he said, the two nations will tie AI to:
- Semiconductors—“because there is no AI without compute.”
- Space—“because the high ground of the modern economy is orbital.”
- Robotics—“because the future of manufacturing is automation married to precision.”
- Energy—“because energy is the base layer of every industrial strategy.”
- “Pax Silica”—the State Department’s flagship effort on AI and supply-chain security—“is a recognition that the map of power has changed,” Helberg said. “And it is our decision to shape that map instead of being shaped by it.”
Also worthwhile:
- ChatGPT’s healthcare push has a fatal flaw … Notably absent from the recent flurry of announcements is Google (Bloomberg)
- Job seekers find a new source of income: Training AI to do their old jobs … Buzzy AI startup employs tens of thousands of white-collar contractors in gigs open to anyone with expertise in their own field (The Wall Street Journal)
From HealthExec’s sibling news outlets:
- Google rolls out 2 new open access AI models aimed at medical imaging (Radiology Business)
- FDA clears opportunistic AI for detecting cardiothoracic issues in CT scans (Cardiovascular Business)
Dave P. has worked in journalism, marketing and public relations for more than 30 years, frequently concentrating on hospitals, healthcare technology and Catholic communications. He has also specialized in fundraising communications, ghostwriting for CEOs of local, national and global charities, nonprofits and foundations.