Brittany Trang, Ph.D., specializes in exploring the intersection of artificial intelligence and healthcare. She delves into questions such as its effectiveness, the beneficiaries, and those who might face disadvantages. Each week, she shares insights in the AI Prognosis newsletter. You can connect with her on Threads, Mastodon, and Bluesky, or reach her via Signal at btrang.01.

At Stanford University, a hub known for its technological innovations—including the invention of the computer mouse and the founding of Google—the focus has shifted towards developing a variety of artificial intelligence tools tailored for the healthcare sector.

However, for the past 18 months, Stanford’s medical center has been prioritizing patient feedback regarding new AI technologies before implementing them.

One of the individuals selected for this initiative is Eric Gries, who has been actively involved as a member of a “patient panel.” Gries has personal experience in caregiving, having supported his wife during her time on a left ventricular assist device (LVAD) and throughout her heart transplant. He also took on the role of temporary caregiver for his brother-in-law during his own heart transplant journey.

Stanford patients reveal challenges in the adoption of health AI.

An airy hospital corridor bathed in warm morning sunlight, where a group of adult patients in masks and white coats cluster around a giant digital health interface screen split by glowing “fault line” cracks. They point and confer anxiously. A red Stanford “S” flag hangs on the wall behind them, casting a soft glow across polished floors.

How Stanford patients help expose ‘fault lines’ in health AI adoption - statnews.com

- Stanford University is developing various AI tools specifically for health care applications.  
- The hospital has consulted patients about new AI tools before their implementation.  
- Eric Gries, a caregiver, participates in Stanford’s patient panel for AI evaluations.

Brittany Trang, Ph.D., covers AI in health and medicine: Does it actually work? Who benefits, or might be harmed? She writes the weekly AI Prognosis newsletter . Follow her on Threads , Mastodon , and Bluesky . You can reach Brittany on Signal at btrang.01.

At Stanford University, it’s easy to get carried away with technology. The computer mouse was invented there. So was Google. And now, it’s pumping out a myriad of tools for artificial intelligence in health care.

But for the last year and a half, Stanford’s hospital has been asking patients about new AI tools before they roll them out.

Advertisement

Eric Gries is one of those people, handpicked by Stanford as part of a “patient panel.” Gries was the caregiver for his wife while she was first on a left ventricular assist device (LVAD), then had a heart transplant. He later became the temporary caregiver for his brother-in-law when he also had a heart transplant.

How Stanford patients help expose ‘fault lines’ in health AI adoption

Sign in to engage with videos and participate in the excitement.

Dr. Peter A. Najjar describes AI in healthcare as 'early stage' yet 'very promising'

A bright hospital corridor flooded with warm morning sunlight streaming through large windows. Dr. Peter A. Najjar, in a crisp white lab coat embroidered with his name, stands holding a sleek tablet that displays a glowing 3D heart model overlaid with dynamic AI-driven analytics charts. His face lights up with excitement as he points to the screen. In soft focus behind him, modern medical monitors and equipment hint at the cutting-edge setting—a hopeful glimpse of healthcare AI’s early days.

Dr. Peter A. Najjar says AI in healthcare is ‘early days’ but ‘very exciting’ - Fox Business

- Users can now log in to comment on videos directly.  
- The new feature promotes greater user engagement and interaction.  
- Joining the fun includes sharing thoughts and insights with others.  

Log in to comment on videos and join in on the fun.

Dr. Peter A. Najjar says AI in healthcare is ‘early days’ but ‘very exciting’

The Wisconsin Department of Natural Resources has issued an air quality warning affecting several counties, including Adams, Buffalo, Clark, Crawford, Grant, Jackson, Juneau, La Crosse, Monroe, Richland, Taylor, Trempealeau, and Vernon. The air quality index (AQI) for fine particulate matter (PM2.5) is anticipated to escalate into the Very Unhealthy (Purple) to Unhealthy (Red) categories statewide. This advisory will remain in effect until noon on Monday.

For those in sensitive groups facing a Very Unhealthy AQI, it is advised to refrain from any outdoor physical activities. Consider rescheduling such activities for when air quality improves or moving them indoors. Individuals with asthma are urged to adhere to their asthma management plans and to keep fast-acting medication readily available. Those with heart conditions should be vigilant for any symptoms such as palpitations, shortness of breath, or unusual fatigue, and should seek medical advice if these occur. The general public is encouraged to minimize outdoor physical exertion and to retreat indoors if they experience any symptoms.

In the case of an Unhealthy AQI, sensitive individuals should also contemplate moving activities indoors or postpone them until air quality improves. People with asthma should continue to follow their action plans and have quick-relief medications accessible. Anyone with heart issues should remain vigilant for adverse symptoms and contact healthcare providers if necessary. Lastly, the broader population should aim to limit the duration and intensity of outdoor activities and seek shelter if they notice any health symptoms.

The current conditions point to another wave of smoke originating from the north-northeast, expected to impact Wisconsin throughout Saturday, with effects persisting into Sunday. Most regions are predicted to experience an Unhealthy AQI, particularly in the northeast where Very Unhealthy levels may occur. Conversely, swaths of the southwestern region may only reach an Unhealthy for Sensitive Groups (Orange) AQI. It is worth noting that this smoke may not reach the southwestern parts of the state until after sunset on Saturday, resulting in statewide air quality effects expected by 6 a.m. on Sunday. For the latest updates on air quality, visit: https://airquality.wi.gov.

A leading global hospital is exploring the transformative potential of AI in health care.

A bright, sunlit hospital research wing with floor-to-ceiling windows, where a doctor in crisp white scrubs studies a large touchscreen display of 3D medical scans feeding from a sleek, AI diagnostic unit beside a patient bed—natural light pouring in to highlight cutting-edge innovation and hope in health care.

One of the world’s most prominent hospitals is testing how AI can revolutionize health care - WXOW

- Wisconsin issues Air Quality Advisory for multiple counties, effective until noon Monday.  
- PM2.5 levels are expected to reach Very Unhealthy and Unhealthy classifications statewide.  
- Residents urged to minimize outdoor activities, especially those with asthma or heart disease.

Weather Alert

The Wisconsin Department of Natural Resources has issued an Air Quality Advisory for the following counties: Adams, Buffalo, Clark, Crawford, Grant, Jackson, Juneau, La Crosse, Monroe, Richland, Taylor, Trempealeau, Vernon WHAT...The PM2.5 AQI is expected to reach the Very Unhealthy (Purple) to Unhealthy (Red) level. WHERE...Statewide. WHEN...Until noon Monday. Precautionary/Preparedness Actions: For Very Unhealthy (Purple) AQI: Sensitive groups: Avoid all physical activity outdoors. Reschedule to a time when air quality is better or move activities indoors. People with asthma: Follow your asthma action plan and keep quick-relief medicine handy. People with heart disease: Symptoms such as palpitations, shortness of breath, or unusual fatigue may indicate a serious problem. If you have any of these, contact your health care provider. Everyone else: Limit outdoor physical activity. Go indoors if you have symptoms. For Unhealthy (Red) AQI: Sensitive groups: Consider rescheduling or moving all activities inside. Go inside if you have symptoms. People with asthma: Follow your asthma action plan and keep quick-relief medicine handy. People with heart disease: Symptoms such as palpitations, shortness of breath, or unusual fatigue may indicate a serious problem. If you have any of these, contact your health care provider. Everyone else: Keep outdoor activities shorter and less intense. Go inside if you have symptoms. Additional Details: Another round of smoke will move in from the north-northeast throughout the day on Saturday before settling over the state on Sunday. Most areas are expected to observe Unhealthy (Red) AQI PM2.5, with the heaviest impacts to the northeast where Very Unhealthy (Purple) AQI PM2.5 will be possible. Some areas to the southwest may only experience Unhealthy for Sensitive Groups (Orange) AQI PM2.5. This round of smoke may not reach the southwestern half of the state before sunset Saturday, meaning the smoke will arrive overnight, with statewide impacts expected by 6 a.m. Sunday morning. For more information on current air quality, please see: https://airquality.wi.gov

One of the world’s most prominent hospitals is testing how AI can revolutionize health care

Mark Cuban responded skeptically to claims about artificial intelligence outperforming human doctors, highlighting a significant discrepancy between the optimistic portrayals of the tech industry and the actual experiences of patients. A survey conducted by the West Health-Gallup Center on Healthcare in America revealed that about 25% of U.S. adults have consulted AI for health-related inquiries.

On July 11, OpenAI CEO Sam Altman shared these survey results, which were quickly endorsed by venture capitalist Marc Andreessen, who remarked that AI is now a superior doctor compared to 99.99% of human medical professionals.

The comparison was drawn from scenarios that had previously puzzled earlier OpenAI models. The results indicated that doctors rated AI's responses as flawless between 19.8% and 23.9% of the time, while human-written answers scored a mere 10.4%.

On July 10, Karan Singhal, head of OpenAI's health research team, released findings from the latest model iteration, GPT-5.6. In blind assessments, healthcare professionals evaluated this AI’s performance on various patient-focused and clinician-focused tasks.

Cuban’s retort on the social media platform X pointed out that over 25% of a physician's time is consumed by dealing with healthcare conglomerates that complicate both the care process and cost-effectiveness for providers and patients alike. His concern isn't about the AI's performance per se, but rather the systemic issues in healthcare that persist even after an accurate diagnosis is made, particularly regarding how insurance companies utilize AI to stall claim approvals instead of facilitating timely care.

On July 12, Andreessen celebrated the AI model's higher performance in a blind test, echoing Altman's claims while also pointing out the detrimental role that intermediaries play in healthcare. He illustrated that these entities are creating AI systems intended to "delay and deny."

In response to the increasing reliance on AI, Cuban noted that many patients remain skeptical of its accuracy, with only 4% of users expressing strong trust in AI's reliability, and 11% reporting that AI provided advice they deemed unsafe. The disconnect lies in patients' apprehensions about the motivations behind AI's guidance.

Cuban’s argument essentially underscores the fact that even if AI tools can optimize certain processes, they don’t address the foundational challenges posed by insurance companies and related entities that prioritize profits over patient care. He drew attention to the financial burden placed on hospitals, which often outsource revenue cycle management, profiting from the delay in claim recovery.

He pointed to his investment in Claimable, a startup leveraging AI to draft appeal letters for patients with denied insurance claims. Claimable, which has secured $10 million in funding, boasts a success rate where approximately 75% of appeals result in a favorable outcome for the patient, illustrating the perpetual confrontation between patient advocacy through technology and insurer resistance.

Unfortunately, many patients remain unaware of this appeal process. According to the KFF, less than 1% of denied claims from Affordable Care Act marketplace plans were challenged in 2023. Of those who did contest their denials, insurers upheld 56% of the rejections, reaffirming the issues Cuban describes.

Cuban’s proposed solution centers on enhancing pricing transparency within the healthcare system. He highlighted the lack of understanding employers have regarding their healthcare expenditures, advocating for direct contracting between employers and healthcare providers. By sidestepping middlemen and negotiating terms directly, he aims to streamline the healthcare process.

To this end, Cuban launched Cost Plus Wellness, an initiative designed to divulge contracts between employers and providers, fostering an environment where others can replicate effective agreements. To date, they have made 37 contracts public, benefitting over 9,000 providers, particularly within the Dallas-Fort Worth area.

Cuban argues that a technologically advanced AI does not inherently lower healthcare costs. Even if AI excels in diagnosis, patients still face bureaucratic hurdles such as prior authorizations and formularies before receiving treatment.

He concluded his message with a stark prediction: unless the middlemen are eliminated from the healthcare equation, the system will remain fundamentally flawed.

Mark Cuban cautions that AI in healthcare will only assist insurers in 'delaying and denying' — responding to Andreessen's assertion about AI doctors.

A dark, cinematic insurance claims office at night: rows of shadowed desks with mechanical arms stamping bold red “DELAY” and “DENY” on medical claim folders. A sleek chrome robot in a lab coat holds a clipboard labeled “AI Healthcare,” standing impassively as claims pile up. In the background, a large screen bathed in warm spotlight shows Mark Cuban’s silhouette pointing defiantly, warning against the system. Deep shadows, moody spotlights, and a tense, film-noir atmosphere.

Mark Cuban warns AI healthcare will just help insurers 'delay and deny' — firing back at Andreessen's AI doctor claim - Yahoo Finance Singapore

- Mark Cuban argues AI's improved diagnosis doesn't address systemic healthcare insurance issues.  
- Survey shows only 4% of users strongly trust AI's medical accuracy, raising patient concerns.  
- Cuban advocates for direct contracting between employers and providers to bypass insurance middlemen.

This led Cuban to reply with skepticism the same day. His perspective reflects a massive gap between tech industry hype and actual patient reality. A West Health-Gallup Center on Healthcare in America survey (5)released April 15 found that one in four U.S. adults has turned to AI at some point with a health question (5).

OpenAI CEO Sam Altman (3)reposted (3) the findings on July 11. Andreessen (4)amplified it (4) the day after with a one-line verdict: "AI is already a better doctor than 99.99% of human doctors."

And these were cases picked specifically because they had stumped earlier OpenAI models. The AI came out ahead by a wide margin: Physicians rated its answers as flawless between 19.8% and 23.9% of the time, depending on the variant, compared with just 10.4% for the physician-written responses (2).

On July 10, Karan Singhal, who leads OpenAI's health research team, (2)posted results from the company's newest model family, GPT-5.6. In blinded reviews, physicians rated GPT-5.6's answers to a set of patient-facing and clinician-facing health tasks.

"25 pct or more of a doctor's time is spent dealing with conglomerates that do all they can to make the doctor's care more difficult and expensive, for both the doctor and patient," Cuban replied in a quote on (1)X (1).

Cuban's rejection wasn't against the score itself, but was aimed at what the two were celebrating, because of what happens after a good diagnosis reaches an insurance company, where the same industry is already using AI to slow claims down rather than pay them out.

On July 12, Marc Andreessen, a venture capitalist, was celebrating an AI model that scored better than real doctors on a blind test, with a post quoting Sam Altman's similar assertion.

Then he went straight for the businesses sitting between patients and their doctors, which, he warned, are building AI agents designed to "delay and deny (1)."

Millionaires under 43 hold only 25% of their wealth in stocks. Here's where their money is actually going

Dave Ramsey warns nearly 50% of Americans are making 1 big Social Security mistake — here's what it is and 3 simple steps to fix it ASAP

Jeff Bezos backs a platform that lets anyone invest in rental homes for as little as $100 — 6 ways to build wealth like a landlord without actually being one

Mark Cuban read a post on X claiming artificial intelligence had already become a better doctor than almost every physician in the country.

Story continues

Among those who had used one in the past 30 days, only 4% said they strongly trust its accuracy and 11% said an AI tool had given them advice they believed was unsafe. Cuban's reply landed in that gap because he knew that in the real world, patients aren't just worried about whether the AI is smart enough; they're worried about who is pulling its strings.

Read More: Are you paying too much for car insurance? Here are 3 clever ways to slash your monthly bill

Why Cuban says the benchmark misses the point

Cuban's argument starts with a number of his own: He says a quarter or more of a doctor's working day already goes to fighting insurers, pharmacy benefit managers and other intermediaries, not treating patients. Handing doctors a faster AI assistant doesn't remove that friction, he argued. It just guarantees the businesses on the other side build faster AI of their own, tools designed to slow down and reject claims rather than approve them (1).

Cuban says this is already happening. Hospitals, he says, pay outside revenue cycle management firms — companies whose whole job is chasing down the money insurers owe — as much as 10% of revenue for the service. He also likened where this is headed to the old Mad magazine cartoon where two spies keep building sharper weapons against each other and neither one ever wins (1).

Actually, there's a real-world test case sitting in Cuban's own investor portfolio. He's an investor in Claimable, a startup that uses AI to write appeal letters for patients whose insurance claims get denied. The startup (6)counts Cuban among the backers behind its $10 million in funding (6), according to Inc., and (7)Bloomberg reported (7) that roughly three out of every four people who appeal through it win a reversal.

In other words, patients are arming themselves with AI appeals, and insurers have every reason to answer with AI denials — Cuban's warning, running in miniature.

And most people never get that far. On Affordable Care Act marketplace plans, for instance, not even 1 in 100 denied claims was challenged by patients in 2023 (8) — and among those who did push back, insurers stood by 56% of the denials, according to KFF, the nonpartisan health policy research group that analyzed the federal claims data.

Bloomberg's reporting on Claimable put a face on that math. One user, a patient managing a chronic inflammatory condition, had relied on the same infusion drug for more than half his life. When his employer switched insurers, the new carrier pushed him toward a cheaper alternative, and a single missed dose during the dispute was enough to trigger a painful flare-up (7).

The medicine worked fine on its own for years. What broke down was a coverage decision made by someone with no stake in his outcome — precisely the friction Cuban is describing.

What Cuban wants employers to do instead

Cuban's fight is over pricing information. He points out in the X post that not a single company, including those run by executives arguing with him online, actually knows what it's paying for the health care it buys its own employees. They see a total bill, but the contract behind that bill stays hidden (1).

To fix this problem, he suggests direct contracting, whereby employers negotiate and pay providers directly, with no carrier. That is, you, as an employer, should avoid third-party administrators, pharmacy benefit managers or any middleman who decides which drugs a plan covers and at what price. He has put money behind the idea.

Cuban's company runs Cost Plus Wellness, a free platform (9)launched in January (9) that publishes contracts between employers and providers so other companies can copy the terms. So far, (10)37 of those contracts are public on the site, reaching more than 9,000 providers (10), according to Becker's — most of them clustered in the Dallas-Fort Worth market, where Baylor Scott & White Health has signed on.

Why a better AI doctor won't lower your bill

For anyone whose premiums climb every January, the benchmark charts miss where the money actually goes. An AI can nail the diagnosis on the first try, and the treatment still has to clear a prior authorization (the insurer's sign-off before care is approved), a formulary (its list of covered drugs) and an appeals queue before anyone gets better.

Cuban ended his post with a prediction blunter than any benchmark. Until the conglomerates are cut out of the middle, he wrote, health care in this country "will continue to be f—ed (1)."

You May Also Like

Here are the 7 top habits of 'quietly wealthy' Americans. How many do you follow?

The tax breaks in Trump's 'big beautiful bill' expire after 2028 — and experts say most people won't act in time. What to do before the window closes

When he dies, Warren Buffett said 90% of his wife's inheritance will go into a single investment. Here's why (and how you can do it too)

Here's the average income of Americans by age in 2026. Are you keeping up or falling behind?

Join 250,000+ readers and get Moneywise's best stories and exclusive interviews first — clear insights curated and delivered weekly. Subscribe now.

Article Sources

We rely only on vetted sources and credible third-party reporting. For details, see our ethics and guidelines.

X (1), (2), (3), (4); Gallup (5); Inc. (6); Bloomberg (7); KFF (8); MedCity News (9); Becker's Health IT (10)

This article originally appeared on Moneywise.com under the title: Mark Cuban warns AI healthcare will just help insurers 'delay and deny' — firing back at Andreessen's AI doctor claim

This article provides information only and should not be construed as advice. It is provided without warranty of any kind.

Mark Cuban warns AI healthcare will just help insurers 'delay and deny' — firing back at Andreessen's AI doctor claim

The ongoing conflict between healthcare providers and insurers has evolved into a sophisticated contest dominated by algorithms, where AI-driven tools for prior authorizations face off against those designed for appeals.

Ashis Barad has firsthand experience from both perspectives of this struggle. He began his professional journey as a physician at Sutter Health and later worked at Baylor Scott & White Health, where he was instrumental in advocating for advanced digital solutions. In 2022, Barad took on the role of chief digital and information officer at Allegheny Health Network and its parent organization, Highmark Health, gaining insights into how insurers develop and implement AI technologies. Two years later, he transitioned back to the provider side by joining the Hospital for Special Surgery (HSS) in the same executive role.

Barad asserts that the ongoing race to automate authorizations and appeals is ultimately a misguided approach. 

“We're engaged in a limited competition while ignoring the broader, ongoing challenge,” Barad pointed out. “If we continue to focus solely on the immediate concerns, we will persist on an inflationary trajectory. We must collaborate to determine how AI can lead us towards a more efficient, cost-effective model because that is what our patients deserve.”

Rather than funneling resources into AI for swifter authorizations and appeals, Barad envisions a future where providers and payers share data to create more individualized care protocols.

He highlighted HSS as a prime example, noting that the institution conducts over 40,000 orthopedic surgeries annually, surpassing any other hospital in the United States. HSS has successfully created a comprehensive database that links imaging results with surgical outcomes.

Such extensive real-world data is a significant gap in the information available to insurers. Barad believes that if payers and providers collaborated to develop these models, authorization processes could become more precise and tailored to individual patients, rather than relying on generic guidelines.

He has already tested this concept in discussions with payers. During a recent meeting with a prominent insurance provider, he inquired about the potential value of HSS's integrated imaging and outcomes data for their actuarial teams.

“They expressed a keen interest, saying they'd find it extremely valuable,” he shared.

Payers generally have access only to basic claims information, such as rates of infection or readmission following initial surgeries, but they lack data on outcomes for patients undergoing multiple procedures. This lack of detail is significant, especially as many HSS patients are complex cases that fall outside typical metrics.

Barad believes that a transformative opportunity exists in utilizing this richer dataset to fundamentally change authorization processes. He pointed to the practice of “gold carding,” which allows exceptional providers to bypass prior authorizations, as an area ripe for reimagining.

Currently, this practice relies on rudimentary criteria, such as flowcharts and cost limits, rather than actual patient outcomes. Barad envisions a redesigned approach personalized for each patient, integrating authorization seamlessly into care pathways rather than treating it as a separate, post-service request.

However, he warns that as long as providers and payers continue to invest in AI merely to outsmart one another, this competition adds unnecessary costs to an already strained system. He emphasizes that the road to genuine cost reduction lies in cooperation and shared data initiatives.

The conflict between providers and payers over bots is increasing costs for all.

Dimly lit industrial hall with two towering mechanical robots locked in a fierce tug-of-war over a giant, cracked dollar coin. One robot bears a glowing medical caduceus emblem, the other a bold insurance-shield icon. Sparks fly from frayed cables as warm orange light flickers off metal surfaces. In the foreground, silhouettes of worried adults clutching empty wallets watch helplessly under heavy shadows.

The ‘Bot Vs. Bot’ Dynamic Between Providers & Payers Is Driving Up Costs for Everyone - MedCity News

- The conflict between healthcare providers and payers revolves around competing AI tools and strategies.  
- Ashis Barad advocates for collaboration on data to create personalized care pathways.  
- Current autorization methods need to evolve from blanket policies to individualized patient approaches.

The war between healthcare providers and payers has increasingly become a battle of algorithms, with AI tools for prior authorizations on one side and AI-powered appeals on the other.

Ashis Barad has sat on both sides of that fight. He began his career as a physician at Sutter Health, then practiced at Baylor Scott & White Health, where he led the push for better digital tools. In 2022, Barad transitioned into the role of chief digital and information officer at Allegheny Health Network and its parent company Highmark Health, getting a taste of how payers build and deploy AI internally. Two years ago, he joined the Hospital for Special Surgery (HSS) to serve as chief digital and information officer, sitting firmly on the provider side once again.

In his eyes, providers and payers racing to out-automate each other is a losing strategy.

“We’re playing a finite game when there’s an infinite game to be played,” Barad declared. “If we play the finite game of today, we’re going to continue on the inflationary route. So we have to figure out how to work together to figure out what the deflationary path for AI is, because that’s what we owe the people, honestly.”

Instead of pouring resources into AI for faster authorizations and appeals, Barad wants to see providers and payers pool their data to build more personalized care pathways.

He pointed to HSS’ own data as an example of what’s possible. The hospital performs more than 40,000 orthopedic surgeries a year — more than double the volume of any other U.S. hospital — and has developed a structured repository connecting imaging data to surgical outcomes.

Having that kind of real-world data is something insurers lack. Barad thinks that if payers and providers built those models together, authorizations would become far more targeted at the individual level, rather than relying on blanket rules.

Barad said he has already tested the idea with payers directly. During a recent meeting with a large insurance company, he asked how valuable HSS’ linked imaging-and-outcomes data would be to their actuarial teams.

“They were like, we would salivate over it,” he said.

That’s because payers typically only have access to basic claims data — infection or readmission rates after a first surgery, for instance — not the outcomes data for patients on their third, fourth or fifth procedure. But those are the kind of complex cases that make up much of HSS’ patient population, Barad pointed out.

Barad thinks the real opportunity lies in using richer data to reshape how authorization works in the first place. He pointed to gold carding, the practice of exempting high-performing providers from prior authorization, as a concept ripe for reinvention.

Today’s version is a blunt instrument based on flowcharts and cost thresholds rather than actual outcomes, he explained. Built the right way, Barad thinks it could instead be personalized down to each patient. Authorization would be baked into a care pathway, not requested after the fact.

But as long as payers and providers keep investing in AI to outmaneuver each other, Barad warned this spending only adds a new layer of cost to a system already straining under it. He said the path to actually lowering costs runs through collaboration on data.

Photo: Malorny, Getty Images

The ‘Bot Vs. Bot’ Dynamic Between Providers & Payers Is Driving Up Costs for Everyone

The global market for AI in healthcare is projected to soar to $491 billion by 2032. However, the World Economic Forum (WEF) issues a cautionary note about potentially repeating the errors of past digital transformations. 

One significant example highlighted is the implementation of electronic health records (EHRs). While EHRs were expected to enhance coordination and elevate clinical decision-making, their introduction initially led to an increase in administrative burdens. This shift often resulted in healthcare professionals spending more time on documentation than on patient care.

The WEF suggests that rather than becoming the focal point of clinician-patient interactions, AI should function discreetly in the background. The aim should be to streamline healthcare processes, thereby enabling clinicians to dedicate more time to developing trust with their patients.

The key takeaway, according to the WEF, is not that digital technologies are inherently flawed, but rather that innovation lacking a clear focus on human-centered objectives can lead to unforeseen challenges.

Building trust between patients and healthcare providers has been a longstanding concern for the WEF, which has recognized it as essential for the effective integration of AI in the healthcare sector. Shyam Bishen, who heads the Centre for Health and Healthcare at the WEF, emphasized in a 2024 article that “mistrust is the most formidable barrier to adopting generative AI in healthcare.”

WEF: AI Could Enhance the Humanity of Healthcare

A bright, sunlit WEF conference stage with a banner reading “WEF” overhead, a doctor gently resting a hand on an adult patient’s shoulder as an AI-powered tablet beside them glows with soft, warm-toned network graphics, warm natural light illuminating their hopeful expressions.

WEF: AI Has the Potential to Make Healthcare More Human - Healthcare Digital

- The global AI healthcare market is projected to reach US$491bn by 2032.  
- WEF warns against past mistakes, citing problematic electronic health records rollout consequences.  
- Trust is crucial for AI adoption; mistrust is a barrier in healthcare settings.  

With the global AI healthcare market on track to reach US$491bn by 2032, the World Economic Forum (WEF) warns against repeating mistakes made during previous waves of digital transformation.

It cites the rollout of electronic health records (EHRs), which promised better coordination and improved clinical decision-making, but initially contributed to growing administrative workloads, often forcing doctors to spend more time documenting care than delivering it.

Instead of taking centre stage in the clinician-patient relationship, WEF argues AI must operate in the background, removing friction from healthcare delivery while allowing clinicians to spend more time building trust with patients.

The lesson, WEF argues, is not that digital technologies fail, but that innovation without a clear human-centred purpose can create unintended consequences.

Creating trust between patient and clinician

WEF has long recognised trust as a critical factor in the successful adoption of AI in healthcare.

“Mistrust is the most formidable barrier to adopting generative AI in healthcare,” Shyam Bishen, Head of the Centre for Health and Healthcare at WEF, wrote in a 2024 article.

WEF: AI Has the Potential to Make Healthcare More Human

The intersection of agentic artificial intelligence (AI) and the healthcare sector has reached a pivotal juncture. With the rapid pace of technological advancement, many healthcare organizations find it challenging to keep up. This delay is not just a matter of efficiency, as it could lead to postponed care, compromised patient privacy, and a breakdown of trust between clinicians and patients.

Leading organizations in this space are typically those that grasp how agentic AI operates, not necessarily the ones that adopt it the quickest. It’s crucial to comprehend where responsible usage of AI can falter.

Exploring the Framework

While agentic AI might seem like a singular entity, it actually functions through five interconnected layers:

1. **Power Requirements**: Similar to all cloud computing frameworks, robust electrical infrastructure is essential to power the AI agents.
  
2. **Infrastructure/Hardware**: Data centers springing up globally are the repositories for the vast amounts of data that AI agents utilize.

3. **Network Layer**: It’s vital to have advanced security measures in place to protect sensitive data.

4. **Data Layer**: The intricate chain of protocols is necessary for effective digital data transmission.

5. **Application Layer**: The various forms in which users engage with AI agents can differ significantly.

Users must understand how these layers interrelate to identify and address issues effectively. Furthermore, reassessing the potential vulnerabilities at each layer is essential for responsible utilization.

Defining Responsible Use

Discussions around responsible AI often focus on the network, data, and application layers, as these are the ones users interact with most frequently. Security breaches can occur even in well-protected organizations if the data layer isn't managed correctly. For example, if an AI learns to equate a patient's pain rating of 10 with "severe pain" without proper context, it can lead to misleading documentation and consequently harm clinician trust and organizational credibility.

Responsible usage of AI demands more than just a secure infrastructure. Three foundational principles must guide any AI program:

1. **Bias Mitigation**: The medical field has historical examples that illustrate the dangers of biased treatments and protocols. AI trained on historical data can replicate these biases on a large scale. Healthcare organizations must consistently review AI systems for biases related to age, race, gender, and other variables, treating this as a clinical necessity rather than a mere compliance task.

2. **Observability**: Transparency is crucial for all personnel interacting with AI agents. They must understand the agent's designated tasks and performance metrics to ensure it functions within established boundaries.

3. **Explainability**: Articulating the functionality and rationale behind an AI agent's actions is essential. Regular third-party audits are important safeguards that enhance accountability both internally and externally.

Effective governance in AI for healthcare requires more than just compliance with a single standard or certification. 

Adopting a comprehensive strategy is vital. Industry benchmarks such as the National Institute of Standards and Technology’s (NIST) Risk Management Framework and insights from the Open Worldwide Application Security Project (OWASP) provide valuable guidance for navigating AI risk and security. Compliance with HIPAA is crucial for every healthcare AI application. Initiatives like the Coalition for Health AI, which I am part of, aim to solidify principles of agentic AI into standardized best practices across the sector.

However, having frameworks isn’t sufficient on its own. The most effective organizations are weaving these standards into their governance models, tailoring them to their specific workflows, patient demographics, and risk assessments, rather than simply fulfilling compliance requirements. The objective is to cultivate trust that elevates AI from a potential liability to a lasting asset.

Thoughtful Implementation Over Hastiness

Healthcare providers cannot ignore the importance of agentic AI, yet they must also refrain from deploying it without adequate infrastructure for responsible use. Finding the balance between overanalyzing—leading to stagnation—and underanalyzing—resulting in reckless deployment—is crucial for strategic success.

Agentic AI serves as a bridge uniting predictive, generative, and autonomous systems. Establishing this bridge sustainably requires active human involvement to set, monitor, and enforce safety parameters over time.

Challenges in healthcare often arise not from the accuracy of AI models, but from determining accountability when errors occur. The most effective adopters create an environment of clinical and organizational trust, transforming agentic AI into a valuable tool for both healthcare teams and their patients.

Steering the Responsible Integration of Agentic AI in Healthcare

A bright, sunlit hospital ward where a female physician in a white coat stands beside a sleek robotic arm offering her a digital tablet displaying a glowing 3D heart scan. A stethoscope and patient charts lie on a nearby stainless steel tray. Warm natural light streams through large windows, casting soft shadows that emphasize human–AI collaboration and responsible oversight in healthcare.

Navigating Responsible Adoption of Agentic AI in Healthcare - HCI Innovation Group

- The healthcare industry faces challenges in adopting rapidly advancing agentic AI technology responsibly.  
- Responsible AI use requires addressing bias, observability, and explainability principles thoroughly.  
- Successful organizations integrate AI frameworks into their governance tailored to specific healthcare contexts.  

The healthcare industry’s relationship to agentic artificial intelligence has reached an inflection point. The technology is advancing faster than most organizations can adopt it — and in healthcare that gap carries consequences for administrators, clinicians, payers and patients. Unlike other industries where delayed adoption means lost efficiency, in healthcare it can mean delayed care, compromised privacy and erosion of clinical trust.

The organizations that lead aren’t necessarily the ones moving fastest, but those who possess a clear understanding of how agentic AI actually works — and where responsible use can break down.

Understanding the Layers

Agentic AI is often thought of as a single system. Functionally, however, it operates across five interdependent layers:

Power requirement: like all cloud-computing platforms, a strong grid is needed to power the computing of every AI agent

Infrastructure/hardware: the data centers that are springing up around the world are where the data every AI agent uses "lives"

Network layer: sophisticated digital safeguards must be in place to ensure private data remains private

Data layer: digital data delivery requires a complex chain of protocols

Application layer: the way a user interfaces with an AI agent can take a variety of forms

Every agentic AI user depends on all five. To troubleshoot a problem, or improve a process, users must know which layer to target. And to use these systems responsibly, we need to understand where each layer is vulnerable.

What ‘Responsible Use’ Really Means

Most conversations about responsible AI focus on the top three layers — network, data and application — because end users interact with these layers most often. Malicious actors can infiltrate an insecure network. But irresponsible use doesn’t require bad intent. A well-secured organization can still fail at the data layer through hallucinations that seem harmless until they aren’t.

Consider a concrete example: A patient rates their pain a 10 on a 10-point scale. An LLM might interpret that rating as “severe pain,” and insert that phrase into a clinical summary, but the physician never used the phrase himself. That single interpretive leap, multiplied across thousands of cases, can create documentation that misrepresents clinical reality, exposes organizations to legal risk, and erodes the trust of the clinicians the system was built to support.

Responsible use requires more than secure infrastructure. In particular, three principles must undergird any AI initiative:

Bias mitigation. The history of medicine offers hard lessons about what happens when treatments and protocols embed the biases of their era. AI systems trained on that same historical data can perpetuate those patterns at scale. Healthcare organizations deploying agentic AI must actively audit for bias across age, race, gender and other factors — not as a compliance exercise, but as a clinical imperative.

Observability. Every person whose work is touched by an AI agent needs visibility into what that agent is designed to do and how it’s performing. Without it, there’s no way to confirm the system is operating within its guardrails — and no way to catch it when it isn’t.

Explainability: The ability to articulate what an AI agent does, and why it does it, shouldn’t be optional. Third-party audits are a necessary safeguard. The capacity to explain a system’s purpose to stakeholders outside the immediate team ensures both internal and external accountability.

Responsible governance in this space — for now, at least — requires more than holding a single membership or certification.

Begin with a multi-pronged approach. Industry frameworks like the National Institute of Standards and Technology’s (NIST) Risk Management Framework and the Open Worldwide Application Security Project (OWASP) offer actionable insights for managing AI risk and security. HIPAA compliance remains non-negotiable for any healthcare AI deployment. Organizations like Coalition for Health AI (whose Agentic AI work group I belong to) are working toward codifying agentic AI principles into industry-wide best practices.

Frameworks alone aren't enough. The organizations already getting this right are integrating these standards into their own internal governance frameworks — contextualizing them to their specific workflows, patient populations and risk profiles — rather than treating compliance as a box to check. The goal is to build the kind of trust that makes AI a durable asset — not a liability.

Deliberate Adoption Beats Moving Quickly

Healthcare organizations can’t afford to ignore agentic AI, but they also can’t afford to deploy it without the infrastructure to use it responsibly. The gap between these two failure modes — overanalyzing to the point of paralysis, and underanalyzing to the point of recklessness — is where the real strategic work lives.

Agentic AI represents a bridge between predictive, generative, and autonomous systems. Building that bridge to last requires keeping a human meaningfully in the loop — the mechanism by which guardrails are set, monitored, and enforced over time.

Health systems struggle less with model accuracy than deciding who is responsible when the model is wrong. The most successful adopters build the kind of clinical and organizational trust that makes agentic AI transformational, both for their teams and for their patients.

Navigating Responsible Adoption of Agentic AI in Healthcare

Just four months prior to Senator Lindsey Graham's passing, I found myself presenting in front of the United States Senate, tasked with addressing a seemingly straightforward query: In what ways can artificial intelligence significantly improve healthcare?

To exemplify my point, I focused on aortic dissection.

“The gravest mistake isn’t a malfunction of the machinery; it’s the oversight or postponement of a diagnosis.”

This past weekend, early reports emerged indicating that Senator Graham succumbed to this very condition.

While the timing is rather unfortunate, I have no insights into Senator Graham's medical treatment, and nothing stated here should imply that alternative technologies would have altered the outcome. Nonetheless, this incident serves to underscore why disorders like aortic dissection are often central to discussions about AI in healthcare.

I could have discussed various conditions such as diabetes or heart failure. However, I chose aortic dissection because it epitomizes one of the toughest challenges within the healthcare landscape. This condition is rare, with symptoms that closely mirror those of more prevalent issues like heart attacks, strokes, severe back, or abdominal discomfort. Accurate recognition requires significant clinical acumen, and once it’s suspected, swift imaging is essential, alongside the immediate mobilization of specialists and often, urgent patient transfers for surgery. Every second is crucial.

Upon reflection, I realized that this example extended beyond just aortic dissection; it was fundamentally about time.

In discussions on healthcare AI, the primary focus is often whether machines can identify diseases better than human clinicians. This is indeed a vital area of progress, but after over two decades of working with healthcare systems, I am convinced that recognition is merely the initial phase of the process.

Throughout my career, I’ve encountered healthcare professionals at various stages of their journey, many grappling with burnout. Never have I met one who intended to overlook a diagnosis. Consequently, I believe we sometimes search for failures in the wrong areas. The healthcare framework is profoundly human-centric. Radiologists conclude their shifts, emergency doctors proceed to the next case, and specialists concentrate on the specific issue at hand. Meanwhile, patients often leave feeling overwhelmed as they juggle personal responsibilities like arranging childcare, caring for aging relatives, or figuring out their work schedules. Information doesn’t typically get lost due to intentional negligence; rather, it becomes obscured amidst the complexities of daily life.

Before my testimony, Dr. Andrew Ibrahim, Chief Clinical Officer at Viz.ai, presented before the House Energy and Commerce Committee to discuss the role of AI in healthcare. His insights resonated with what I hoped lawmakers would grasp. “What truly counts,” he emphasized, “isn’t just the algorithm itself, but how these technologies are embedded within actual clinical workflows to address significant, time-critical issues.”

One of the most impact-full moments from Ibrahim’s testimony was when he shared a personal experience involving his father.

When Ibrahim’s father exhibited stroke symptoms, his medical knowledge enabled him to recognize the situation promptly. He informed the emergency department prior to his father’s arrival and coordinated care that ultimately resulted in a favorable outcome.

In a recent conversation, Ibrahim shared that this experience brought to light an aspect he hadn’t fully grasped even during his medical training: “I didn’t realize how challenging it is for information to reach the right individual,” he remarked. “There’s someone at the end of that communication loop who needs to make a critical decision quickly.”

He further outlined that AI's potential lies not only in spotting potentially life-threatening conditions but also in efficiently transmitting the necessary information and images to the specialists positioned to act.

He made another important observation that extends well beyond stroke cases:

“Not everyone has a medically knowledgeable relative ready to coordinate their emergency care. We must develop systems that guarantee patients receive appropriate treatment without depending on luck or personal connections.”

This is the thought that has lingered with me. We’ve dedicated considerable time questioning whether AI can outperform doctors in disease diagnosis. I’ve come to believe that the true strength of AI lies in its ability to minimize the degree to which exceptional outcomes hinge on mere chance.

Lindsey Graham's Passing Highlights the Importance of Healthcare AI

A dimly lit hospital ICU at night, the single bed empty except for a rumpled gown and a flatlined heart-monitor screen casting ominous red glows. Beside it, a sleek robotic arm in standby, digital patient charts flickering “CRITICAL” alerts. A small U.S. flag hangs quietly on the wall, shadows swallowing the corners—evoking loss and the unheeded warnings of healthcare AI.

What Lindsey Graham’s Death Reminds Us About Healthcare AI - Forbes

- Senator Graham reportedly died from aortic dissection, underscoring AI's potential impact in healthcare.  
- Aortic dissection exemplifies challenges in timely diagnosis and treatment within healthcare systems.  
- Effective healthcare AI should streamline communication and information flow for timely decision-making.  

Close-up of CT scan results reflected in glasses of physician. Universal Images Group via Getty Images

Four months before Senator Lindsey Graham’s death, I stood before the United States Senate trying to answer a deceptively simple question: Where can artificial intelligence make the biggest difference in healthcare?

To illustrate my answer, I chose aortic dissection.

“The most dangerous failure is not a machine failure. It’s a missed or delayed diagnosis.”

This weekend, preliminary reports indicated that Senator Graham died from that very condition.

The timing is an unfortunate coincidence. I have no knowledge of Senator Graham’s medical care, and nothing written here should be interpreted as evidence that different technology would have changed the outcome. But this moment does remind us why conditions like aortic dissection have become defining examples in conversations about healthcare AI.

I could have spoken about diabetes, heart failure or countless other diseases. Instead, I chose aortic dissection because it captures one of healthcare’s hardest challenges. The condition is uncommon. Its symptoms often resemble far more common illnesses, including heart attack, stroke, severe back pain or abdominal pain. Recognition requires clinical expertise, but diagnosis is only the beginning. Once suspected, imaging must be interpreted quickly, specialists mobilized, and patients frequently transferred for emergency surgery. Every minute matters.

MORE FOR YOU

Looking back, this example was never really about aortic dissection; it was about time. At least, that’s what I eventually realized.

Healthcare AI is usually framed as a question of whether computers can recognize disease better than physicians. That is certainly an important area of innovation, but after more than twenty years helping health systems deploy clinical technology, I’ve become convinced that recognition is only the first step.

In more than twenty years working with health systems, I’ve met clinicians at every stage of their careers and every stage of burnout. I’ve never met one who wanted to miss a diagnosis. That’s why I think we sometimes look for failure in the wrong place. Healthcare is an extraordinarily human system. Radiologists finish shifts. Emergency physicians move on to the next trauma. Specialists focus on the problem they were asked to solve. Patients go home overwhelmed, trying to make sense of everything that happened while simultaneously figuring out how they’ll get their children to school the next morning, who will take an aging parent to an appointment, or whether they’ll be able to return to work. Information isn’t usually lost because someone intended for it to disappear. It gets buried beneath the realities of everyday life.

Months before I testified, Dr. Andrew Ibrahim, Chief Clinical Officer at Viz.ai, appeared before the House Energy and Commerce Committee to discuss healthcare AI. His testimony echoed what I hoped lawmakers would understand. “What matters,” he explained, “is not the algorithm in isolation, but how these tools are integrated into real clinical workflows to solve real, high-stakes problems.”

For me, the most memorable part of Ibrahim’s testimony was an experience he shared about his father.

When his father developed stroke symptoms, Ibrahim recognized what was happening because of his clinical knowledge. He alerted the emergency department before his father arrived and coordinated the care that ultimately helped produce a good outcome.

When we spoke recently, Ibrahim told me that the experience revealed something he had not fully appreciated even during his medical training. “I didn’t appreciate how hard it is for information to get to the right person,” he said. “There’s a person at the end of that telephone game who has to make a time-sensitive decision.”

The opportunity for AI, he explained, is not only to recognize a potentially life-threatening condition but to move the relevant images and information through that chain to the specialist who must act.

His testimony made a related point that extends far beyond stroke:

“Not everyone has a medically trained son ready to coordinate their emergency care. We need systems that ensure patients get the right treatment without relying on luck or personal connections.”

That’s the sentence I haven’t been able to move beyond. We’ve spent years asking whether AI can diagnose disease better than physicians. I’ve come to think AI is at its best when it makes extraordinary outcomes less dependent on extraordinary luck.

What Lindsey Graham’s Death Reminds Us About Healthcare AI

As healthcare institutions and their partners embark on the journey of integrating autonomous artificial intelligence and various AI technologies, it's vital for them to perform a comprehensive assessment of their data assets. Understanding the permissible uses of this data under HIPAA and associated regulations is crucial, stated attorney Jordan Cohen from Akerman LLP.

In an insightful discussion with ISMG, Cohen highlighted the implications of straying from legal data usage. He pointed out that if protected health information (PHI) is mishandled, it could result in a reportable breach.

Cohen emphasized that many of the essential strategies HIPAA-compliant entities should adopt during their implementation of autonomous AI are not exclusively tied to AI. A meticulous inventory of data flow is essential. He advised that organizations should map out and account for how data is collected, processed, stored, and transmitted, including how vendors interact with it. "If you lack clarity on what data exists, its location, and who has access to it, securing that data and ensuring patient privacy becomes a formidable challenge," he remarked.

While these data management practices have been promoted for years, their importance has grown exponentially in the era of autonomous AI deployments.

In his video interview with ISMG, Cohen further explored several topics, including:

- The prevalent applications of autonomous AI in clinical and administrative settings within healthcare, along with the types of PHI, electronic health records, and other data being commonly utilized.
- Additional legal and regulatory considerations relevant to the deployment of AI technologies.
- Key considerations for AI, such as implementing technical safeguards, effective incident response strategies, ongoing monitoring, and ensuring transparency and patient consent.
- The potential for AI to enhance data privacy and security measures across healthcare and other industries.

Cohen serves as a partner at Akerman LLP and leads the firm’s digital health practice, where he provides legal guidance to clients regarding transactions involving healthcare providers and related platforms. His expertise encompasses issues of federal and state privacy compliance, including adherence to HIPAA's provisions regarding Privacy, Security, and Breach Notification, along with navigating state-specific breach notification laws. Additionally, he advises on a wide array of regulatory matters pertinent to healthcare, such as compliance with fraud and abuse regulations like the Anti-Kickback Statute and the Stark Law.

The Need for Enhanced Data Oversight in AI Healthcare Applications

A quiet hospital data hub at dawn: a physician in a crisp white coat stands before a sleek workstation where cascading, semi-transparent streams of medical records and AI-generated charts swirl around a glowing red padlock icon, natural morning light filtering through tall windows to underscore the urgent need for stronger data oversight in AI-powered healthcare.

Why AI in Healthcare Demands Stronger Data Oversight - BankInfoSecurity

- Healthcare organizations must inventory data for HIPAA compliance amid AI integration challenges.  
- Failures in data usage can lead to breaches involving protected health information.  
- Critical practices include understanding data flow and safeguarding patient privacy in AI deployments.  

Artificial Intelligence & Machine Learning , HIPAA/HITECH , Next-Generation Technologies & Secure Development

Why AI in Healthcare Demands Stronger Data Oversight

Attorney Jordan Cohen of Akerman on AI Challenges Ahead

Jordan Cohen, partner, Akerman LLP

As healthcare organizations and their vendors develop and implement agentic artificial intelligence and other AI technology, they should conduct a thorough data inventory and have a solid understanding of what data can and cannot be used under HIPAA and other laws, said attorney Jordan Cohen of law firm Akerman LLP.

See Also: Know Thy Enemy: Threats to Cyber Resilience

"If you fall outside of a permissible use, then technically, if protected health information is involved, that can be considered a reportable breach," he said in an interview with ISMG.

Many of the steps that HIPAA-regulated enterprises should take in their agentic AI implementations aren't really "AI specific," he said.

"A data flow inventory is going to be really important. So diagramming and accounting for how you're ingesting data, processing it, storing it and how it's leaving your systems, how vendors are touching it and what they're doing to that data is going to be critical," he said.

"Because if you don't know what data you have, where it lives, who's accessing it, then it's really difficult to secure it and to protect patient privacy," he said.

"These are practices that we've been discussing for years," he said. But in the age of agentic AI and other AI deployments, they are especially important.

In this video interview with ISMG, Cohen also discussed:

Most common current uses of agentic AI for clinical and administrative purposes in healthcare, and the types of PHI, electronic health record and other data frequently being utilized;

Other legal and regulatory issues to consider in the deployment of AI;

Other critical AI issues to consider, including technical safeguards, incident, response, monitoring, transparency and consent of patients;

Opportunities for AI to improve data privacy and security in healthcare and other sectors.

Cohen is a partner at Akerman LLP and team leader of the firm's digital health practice. He advises clients on transactions involving healthcare providers and other healthcare-related platforms. That includes providing legal counsel related to federal and state privacy and data security matters, including compliance with HIPAA's Privacy, Security and Breach Notification Rules, as well as compliance with state breach notification laws. He also provides guidance on a broad range of healthcare regulatory matters, including compliance with fraud and abuse laws including the Anti-Kickback Statute and the Stark Law.

Why AI in Healthcare Demands Stronger Data Oversight

Sign in to share your thoughts on videos and participate in the excitement!

Strategist claims healthcare may emerge as an early AI 'winner.'

Sunlit modern hospital ward with a white-coated physician reviewing a tablet displaying AI-analyzed medical charts and transparent digital organ overlays; on a nearby counter sits a small trophy etched with a caduceus and laurel, symbolizing healthcare’s early AI victory.

Healthcare could become an early AI 'winner,' strategist says - Fox Business

- Users can now log in to comment on videos for enhanced interaction.  
- New feature encourages community engagement and discussion around video content.  
- Users gain access to more personalized content through their login accounts.  

Healthcare could become an early AI 'winner,' strategist says

In Rochester, Minnesota, Mayo Clinic's internal medicine physician, Dr. Alexander Ryu, faces the daunting task of reviewing extensive medical records when preparing to consult with patients. Many individuals seeking treatment at the clinic bring an array of disorganized documents from various health systems, often looking for second or third opinions. To streamline this process, Mayo Clinic has introduced an innovative AI tool named Record Time, which assists clinicians in rapidly analyzing patient records. This tool not only organizes documents chronologically but also generates relevant summaries, making vital information easier to locate.

Dr. Ryu notes that Record Time can reduce his preparatory time for each patient visit by five to thirty minutes, a valuable efficiency that allows him to spend more time directly with patients. The tool is designed to support doctors by highlighting crucial details that could affect treatment decisions and diagnostic recommendations—a significant advantage considering the tens of millions of pages of medical records the clinic processes annually. Ryu, who is also the vice chair of innovation for the Mayo Clinic's Department of Medicine, emphasizes the need for effective solutions to manage the vast volume of patient information.

The potential of AI in healthcare is widely recognized, with major companies like Google, OpenAI, and Anthropic launching health assistant chatbots in response to a growing public interest in AI for medical inquiries. While some tech leaders tout ambitious claims about AI's ability to revolutionize disease treatment, the Mayo Clinic approaches AI more pragmatically, focusing on its capacity to enhance patient care and save lives. In collaboration with organizations such as Microsoft and Scale AI, Mayo Clinic is leveraging its large database of patient records and clinical research to refine AI applications. Currently, around 150 AI models are operational within the institution.

However, the integration of AI in healthcare raises critical concerns about accuracy and patient privacy. Recently, Mayo Clinic's former Director of Research Operations, Traci Tamiko Eto, filed a lawsuit against the hospital, alleging retaliation for voicing concerns regarding privacy and oversight related to some of the AI systems.

A spokesperson for Mayo Clinic, Andrea Kalmanovitz, stated that while the hospital does not comment on ongoing legal matters, it is dedicated to the responsible development of AI, emphasizing the importance of privacy, security, and compliance in all initiatives. She affirmed the institution’s commitment to upholding patient trust and adhering to legal guidelines.

The utility of AI in medicine hinges on its capacity to detect trends within large datasets. Jason Droege, CEO of Scale AI, highlighted how AI can handle the tedious tasks typically performed by healthcare professionals, thereby accelerating the diagnostic process and supporting more effective treatment.

Dr. Matthew Callstrom, a radiologist and leader of Mayo Clinic’s generative AI program, believes strongly in AI’s potential. He cites a clinical trial aimed at assessing whether AI can recognize patients at risk for early-stage pancreatic cancer, a detection method that could lead to interventions years earlier than current practices, which usually diagnose the disease at an advanced stage with a meager five-year survival rate.

Additionally, Mayo Clinic has utilized AI for monitoring patients' cardiac rhythms to predict conditions like atrial fibrillation, which can lead to serious complications such as blood clots or strokes. According to Callstrom, identifying such conditions early can have a transformative impact on patient care.

To develop AI solutions, Mayo Clinic collaborates closely with both technical experts and clinicians to identify priority health challenges. Callstrom emphasizes the rigorous validation process for AI tools, which includes small-scale testing under physician supervision, with expansions based on performance outcomes. Continuous monitoring follows after tools are widely implemented.

Addressing concerns amongst staff regarding the implications of AI for employment, Callstrom reassured that while roles may evolve, they are not disappearing. For instance, nurses have been involved in creating an AI tool that documents patient interactions, significantly reducing the time spent on administrative tasks and enhancing direct patient engagement.

Droege believes the healthcare sector is still in the early stages of unlocking AI’s comprehensive potential. However, he argues that the emphasis should be placed on the quality of care rather than the speed of deployment. As he puts it, achieving optimal outcomes should be the primary goal, and speed should come only after ensuring accuracy and reliability in healthcare applications.

A leading global hospital is experimenting with AI to transform healthcare.

A modern hospital ward flooded with warm natural light as a surgeon and nurse study dynamic 3D patient scans on a sleek touchscreen console, a robotic medical arm poised to hand over surgical tools, the hospital’s grand facade visible through large windows, conveying a new era of AI-driven healthcare.

One of the world’s most prominent hospitals is testing how AI can revolutionize health care - CNN

- Mayo Clinic's AI tool, Record Time, helps organize patient records efficiently for doctors.  
- Over 150 AI models are deployed at Mayo Clinic to enhance patient care and diagnostics.  
- Concerns about AI accuracy and privacy persist, highlighted by a recent lawsuit against the hospital.  

Rochester, Minn. —

Preparing to meet with a patient can require Mayo Clinic internal medicine physician Dr. Alexander Ryu to sort through dozens, or even hundreds, of pages of medical records.

Many patients visit the renowned clinic looking for a third or fourth opinion, carting with them unsorted documents from external health systems. A new artificial intelligence tool is helping clinicians parse through records faster — generating relevant patient summaries, organizing documents in chronological order and making them easier to search.

Ryu said the tool, called Record Time, can save him between five and 30 minutes of preparation per visit, depending on the complexity of the case. That’s time he can instead spend face-to-face with the patient. And Record Time helps ensure that he doesn’t miss important details that might be buried in the file that could drive treatment and testing recommendations.

“We receive a huge volume of these records, tens of millions of pages every year, and we needed a way to find important information in that,” said Ryu, who also serves as vice chair of innovation for the Mayo Clinic Department of Medicine.

Mayo Clinic physician Dr. Alexander Ryu helped to develop an AI system to better prepare doctors for patient visits. CNN

Health-related applications are considered among the most promising areas of innovation for AI. Google, OpenAI, Anthropic and others have rolled out health assistant chatbot features — and tens of millions of people are turning to AI for medical-related questions. Leaders in Silicon Valley also frequently make bold promises about AI curing cancer and other diseases, although those statements often sound more like marketing pitches given the big industry players are largely focused on other consumer and business applications.

But Record Time is just one way that Mayo Clinic, one of the world’s most well-known hospital systems, is using AI in hopes of improving patient care and, ultimately, saving lives. The hospital is partnering with firms like Microsoft and Scale AI to use its huge volume of patient records and research to develop AI tools.

There are now around 150 AI models deployed within the hospital, according to Dr. Matthew Callstrom, a radiologist and medical director of Mayo Clinic’s generative AI program.

The use of AI in healthcare settings is not without controversy, raising big questions about accuracy and patient privacy.

Mayo Clinic’s former Director of Research Operations Traci Tamiko Eto sued the hospital earlier this month, alleging she was retaliated against for raising privacy and oversight concerns around some Mayo AI systems.

Mayo Clinic spokesperson Andrea Kalmanovitz said the hospital doesn’t comment on ongoing litigation but noted it is “committed to the responsible development and deployment of AI, with privacy, security, transparency and compliance embedded throughout our processes.”

“Our research and clinical innovation are conducted in accordance with applicable laws and regulations and we remain steadfast in upholding the trust patients place in us and respecting their privacy,” Kalmanovitz said in a statement.

‘Potentially life changing’

Part of what makes AI so useful in health and medicine is that the technology excels at identifying trends in large swaths of data, said Jason Droege, CEO of Scale AI, which worked with Mayo Clinic to develop Record Time.

“AI can step in and do a lot of the tedious work that very specialized doctors or medical professionals do to speed up that process — get to more accurate diagnoses, faster so you can treat more people,” he said. “This is an industry where a lot of what doctors are doing, and nurses and others, is pattern recognition.”

Callstrom told CNN he was convinced of AI’s potential back in 2016, when he saw how AI could help radiologists identify subtle, early cancer warning signs in imaging.

Mayo Clinic is now running a clinical trial to test whether AI can help identify patients at risk of or with early-stage pancreatic cancer — an application the hospital has said could detect the disease years earlier than the typical diagnosis. Currently, patients often aren’t diagnosed with pancreatic cancer until it is regionally spread or metastatic, when the five-year survival rate hovers around 9%, Callstrom said.

Scale AI CEO Jason Droege discusses the company's partnership with Mayo Clinic to develop AI tools for health care. CNN

The hospital has also successfully used AI to analyze patients’ heart rhythms to tell whether someone could develop atrial fibrillation, a condition that can cause blood clots and strokes.

“For those patients where you actually find it and identify it, it’s potentially life changing,” Callstrom said.

Balancing speed and trust

To build AI tools, Mayo Clinic pairs tech experts with doctors or clinicians to decide which medical problems need tackling. A big part of Callstrom’s job is ensuring that Mayo Clinic’s AI tools are accurate and trusted by both patients and doctors.

AI tools go through the same process as a clinical trial, he said. First, it’s tested with a small group of patients with doctor oversight. The performance is measured, and then testing expands to a wider population.

Once a tool is rolled out broadly, Mayo Clinic continues to monitor how well it works.

“On the physician side … we’re skeptical a lot,” Callstrom said. “We give them the option to try (a new AI tool) and if they like it, they use it. If they don’t want to use it, they don’t have to. And the best measure of how well we’re doing is the adoption rate.”

Dr. Matthew Callstrom says AI shows huge potential for early disease detection, potentially saving patient lives. CNN

Callstrom also gets frequent questions from staff about what AI tools will mean for jobs. So far, he said, jobs aren’t going away. But they are changing.

The hospital’s nursing team, for example, helped to develop an AI system that listens and takes notes during patient visits. That potentially cuts in half the more than an hour per day they spend typing up those visits.

“What it’s doing is letting them spend more time talking to patients,” Callstrom said.

Scale’s Droege believes the health care industry is only at the beginning of realizing AI’s full potential — but that speed of adoption shouldn’t necessarily be the top priority.

“These predictions where everything is going to be fixed in a year or two, I think that’s wildly ambitious,” he said. “Quality of care is the bar, and then speed … in healthcare, you want to get it right, as fast as possible.”

Stanford patients reveal challenges in the adoption of health AI.

More from AI in healthcare

Dr. Peter A. Najjar describes AI in healthcare as 'early stage' yet 'very promising'

A leading global hospital is exploring the transformative potential of AI in health care.

Mark Cuban cautions that AI in healthcare will only assist insurers in 'delaying and denying' — responding to Andreessen's assertion about AI doctors.

The conflict between providers and payers over bots is increasing costs for all.