OpenAI is making a major bet on the healthcare industry, having launched three new healthcare products over the past six months, including an updated version of ChatGPT for clinicians and the “ChatGPT Health” tool. It has already secured eight large healthcare systems as customers, including Cedars-Sinai Medical Center and HCA Healthcare, with over 230 million users worldwide receiving health advice via ChatGPT each week. Healthcare has become one of OpenAI’s core strategic priorities, given that it accounts for approximately 18% of the U.S. economy. However, the company faces competition from Anthropic, Google, and others, reported a net loss of $39 billion in 2024, and must also address challenges related to the accuracy of AI-generated medical advice and potential legal risks.

Article author and source: Forbes (ID: forbes_china)

This AI giant is fully betting on the healthcare industry, having launched three new products in the past six months. But it must ensure ChatGPT’s accuracy is sufficiently reliable, as hundreds of millions of people have already turned to it for healthcare advice.

When OpenAI seeks to pitch its healthcare initiatives to top U.S. healthcare systems, it often invokes a figure hospital CEOs cannot ignore: Sam Altman.

Altman, 41, is the co-founder and CEO of OpenAI, a billionaire who excels at alleviating external concerns and closing deals. He has successfully convinced investors, boards, and governments around the world that OpenAI is the driving force behind the next era of computing. Now, he is personally promoting this vision to hospitals.

01

Altman’s personal involvement in sales discussions underscores the central role of the healthcare business in OpenAI’s overall strategy. In January this year, the AI giant announced that eight major healthcare systems, including Cedars-Sinai Medical Center and HCA Healthcare, have become customers of its enterprise-grade healthcare tools.

OpenAI has also invited hundreds of doctors to refine health-related Q&A content, serving over 230 million users worldwide each week who seek advice through ChatGPT. The company is launching a new version of ChatGPT tailored for clinicians, as well as “ChatGPT Health”—a tool embedded within the main app that allows users to securely link their medical records and health apps, such as Apple Health and MyFitnessPal, though it is currently available only via a waitlist. Additionally, several healthcare technology companies are developing tools based on OpenAI’s models to generate clinical documentation and assist users in interpreting lab results. In just the past six months, OpenAI has launched three new products for the healthcare sector.

“Healthcare is one of OpenAI’s most important business areas,” said Nate Gross, head of OpenAI’s healthcare division. Gross joined OpenAI in 2025 and holds an M.D. from Emory University School of Medicine and an MBA from Harvard Business School. Previously, he co-founded Doximity, a healthcare social platform valued at $4 billion. Although the company is also expanding into other large markets such as education and finance, Gross said, “Everyone encounters health issues—by focusing on healthcare, we have the opportunity to help everyone.”

After all, the healthcare industry accounts for approximately 18% of the U.S. economy. If developed properly, AI holds tremendous potential in this sector: it can help consumers make more informed health decisions, assist doctors in improving diagnostic quality, and enable healthcare institutions to enhance operational efficiency.

Nate Gross, head of OpenAI’s healthcare business. Image credit: OPENAI

02

As OpenAI fully enters the healthcare sector, the company reaches a critical juncture of success or failure.

Three years ago, OpenAI sparked an AI revolution with ChatGPT, but its lead now appears somewhat diminished as competitors like Anthropic and Google rapidly rise. In 2025, the company reported revenue of $13 billion, yet its net loss reportedly surged from $5 billion in 2024 to $39 billion last year—an almost sevenfold increase. In April this year, reports emerged that OpenAI failed to meet key revenue and user targets. This industry giant, valued at $852 billion, secretly filed for an IPO in June, but is reportedly considering delaying the offering until next year amid investor pressure.

Although the company will not separately disclose financial data for its healthcare business prior to listing, the healthcare business is clearly key to its future success.

But OpenAI must outcompete a group of rivals, including Anthropic and Google, which have focused on enterprise services since their inception and possess their own portfolio of healthcare products, as well as all other health tech companies targeting the same market. As large model capabilities become increasingly homogenized, it must also demonstrate price competitiveness—after all, hospitals operate on extremely thin profit margins. “I’m very optimistic about their competitiveness in the consumer market,” said John Beadle, managing partner at Aegis Ventures, which specializes in healthcare. “But I’m less confident about their prospects in the enterprise market.”

In February 2024, Lauren Bannon began feeling unwell. She struggled to bend her fingers, especially in the morning and evening. For the first few months, she dismissed the pain, but eventually it spread to her abdomen. After a series of tests, doctors told her she had rheumatoid arthritis, a condition that cannot be detected through blood tests.

But the 42-year-old marketing agency founder from North Carolina was unconvinced and, like many others, turned to ChatGPT for answers. Bannon asked the AI a series of questions about her symptoms, and the AI suggested she might have Hashimoto’s disease—a condition in which the immune system attacks the thyroid. A thyroid ultrasound confirmed this diagnosis, and the doctor also discovered two small nodules. Bannon was later diagnosed with aggressive thyroid cancer. In January 2025, surgeons removed her thyroid and two lymph nodes where cancer had spread.

"ChatGPT really saved my life. I'm not just saying that—I truly believe it." Bannon said.

Among the millions of users who consult ChatGPT about health-related issues, most do not have experiences like Bannon's. They ask simple questions such as, “Why does my knee hurt when I climb stairs?” or “Which exercises help maintain bone density?” People use it to interpret medical terminology, analyze lab results, and understand drug effects. However, some also report unusual symptoms to it—symptoms that are sometimes later diagnosed as serious conditions.

03

“Today, health is one of ChatGPT’s most active use cases,” said Karan Singhal, Head of Health AI at OpenAI. “It can bring both benefits and harms, but the potential for good is much greater.”

In the United States, a shortage of clinicians combined with excessive administrative burdens has led to physician burnout and made it difficult for patients to schedule necessary care. In rural areas, more than one-third of adults visit emergency rooms for conditions that could be treated in primary care settings; 86% of agricultural counties have no practicing cardiologist at all.

Access to medical resources is a critical issue. Hundreds of millions of users are turning to ChatGPT for essential information—they may have to wait six months to see a doctor, or they may feel terrified and unsure of what to do next,” said Ashley Alexander, Head of Health Products at OpenAI. “We clearly state that ChatGPT cannot replace medical care, but in moments of helplessness and fear, it can provide users with reference information.”

But like the early days of “Dr. Google,” using ChatGPT for health reference or self-diagnosis is a double-edged sword. While this tool can certainly be helpful, it may also provide inaccurate information, leading patients to make unnecessary emergency room visits or overlook symptoms that deserve attention.

A study published in Nature Medicine in February evaluated OpenAI’s new consumer-grade health tool, ChatGPT Health. The study found that in over half of the cases where medical attention was warranted, it failed to recommend that patients seek hospital care. It missed identifying users who expressed suicidal ideation and explicitly mentioned self-harm methods, but it correctly identified users who did not disclose specific self-harm methods as requiring intervention. In non-emergency cases, 65% were advised to seek medical care when it was unnecessary, potentially adding strain to an already overburdened healthcare system. OpenAI responded point by point on X, with Gross noting, “This study has several obvious methodological flaws and used an outdated version of the model.”

Ashley Alexander, Head of Health Products at OpenAI. Image source: OPENAI

It has been reported that OpenAI has received subpoenas from several U.S. states investigating issues related to health data handling and security. A Canadian woman has sued the company, claiming that her 24-year-old daughter died by suicide and that ChatGPT played a role in encouraging the act. OpenAI stated that it “takes seriously” the concerns raised by state attorneys general and will respond “constructively.”

04

Doctors have personally experienced the advantages and disadvantages.

Dr. Jinsy Andrews, a neurologist specializing in ALS at NYU Langone Health, says patients frequently come to appointments with printed ChatGPT chat logs. She notes that while ChatGPT can sometimes encourage people to seek medical care earlier and receive earlier diagnoses for ALS—a progressive, incurable neurological disease—it can also cause unnecessary distress. One patient worried they had ALS because ChatGPT suggested muscle twitching might be a sign. Andrews clarifies that this is not accurate, but it “increased the patient’s anxiety.” Another patient requested trying an experimental therapy recommended by ChatGPT, only to discover that the cited research was entirely fabricated. “It takes time to correct and clarify these misinformed searches, which detracts from the time needed to thoroughly evaluate the actual condition and discuss treatment options,” she says.

An anonymous doctor at the Albany Medical Health System shared several cases in which patients cited ChatGPT’s advice to challenge their doctor’s diagnosis and propose treatment plans with no scientific basis. One patient, after inputting his symptoms into ChatGPT, was told he might have an autoimmune disease. “He refused to undergo any further tests to confirm the diagnosis and insisted, ‘This is definitely what I have. ChatGPT said so, and this is how I should be treated,’ ” the doctor said.

OpenAI is well aware of these risks and has assembled a team of over 260 doctors to refine the health advice provided by ChatGPT. These doctors evaluate ChatGPT’s responses, discuss specific cases, and, most importantly, train the model to refrain from giving recommendations when information is insufficient.

“When I first joined, I thought its performance wasn’t great,” says Rebecca Soskin Hicks, a pediatrician trained at Stanford who was hired by OpenAI about two and a half years ago to conduct red team testing on the model and now leads the company’s physician network. “To be honest, it wasn’t hard to trick it into making mistakes back then.”

Since then, the team has reviewed over 700,000 response samples, significantly improving the model’s accuracy when answering questions from general users and clinicians. OpenAI states that millions of clinicians now use ChatGPT each week. OpenAI’s proprietary benchmarking tool, HealthBench, evaluates model performance across 48,000 criteria, including triage and appropriately expressing uncertainty when needed. Results show that OpenAI’s new model demonstrates significant improvements over previous versions in medical tasks, but still falls short in proactively asking for missing information.

05

However, in medical operation scenarios, OpenAI's models are more reliable.

AdventHealth, headquartered in Florida and operating more than 50 hospitals, reported an 80% reduction in time spent on administrative tasks after implementing OpenAI models. The Memorial Sloan Kettering Cancer Center in New York has launched a pilot program involving 1,000 participants, but it is still too early to assess the actual impact of OpenAI. “Initial feedback has been positive,” said Ophelia Chiu, Vice President of Strategic Innovation at the center. “Despite the heavy marketing hype over the past three years, healthcare institutions are still in the very early stages of exploring AI applications.”

Gross from OpenAI said that many of the company’s visions in the healthcare field will be realized through startups and enterprises developing new products based on its models.

Take Labcorp, a diagnostics giant with a market capitalization of $23 billion, as an example. A recent survey by the company found that 41% of people are already using AI to interpret lab reports, but the accuracy of information varies widely across different tools. In response, in May of this year, Labcorp partnered with OpenAI to launch a new mobile app powered by AI, designed to help users understand their lab results and track long-term trends in key metrics.

“We brought in doctors to dig into it and find its flaws, but its accuracy turned out to be very high,” said Bola Oyegunwa, Chief Information and Technology Officer at Labcorp.

Thousands of other institutions and startups, such as Abridge, Ambience, and EliseAI, are also using OpenAI’s models.

Taking Abridge as an example, the company is best known for its technology that automatically records doctor-patient conversations and generates clinical notes. Davis Liang, head of the company’s machine learning team, said Abridge combines cutting-edge large models from OpenAI and Anthropic with its own proprietary models to deliver an optimal product for clinicians. OpenAI’s models excel at note-taking and general user tasks, but perform slightly weaker in evaluating patient symptoms and generating diagnostic codes—which are also used for billing. “This task is extremely challenging due to the high demand for precision,” he said. “OpenAI might simply identify that a patient has diabetes, when in reality the patient has diabetes with diabetic retinopathy.”

Nevertheless, AI technology is advancing at an extremely rapid pace, making it difficult to predict what this field will look like in six months, let alone five years.

“There’s no silver bullet to move an industry that accounts for 20% of GDP,” says Gross at OpenAI. “All we can do is ensure our models genuinely help people.”