OpenAI and Anthropic Bet Big on Health Care AI

According to Bloomberg Business, OpenAI announced on Thursday a version of ChatGPT designed for clinicians at institutions like Boston Children’s Hospital and Memorial Sloan Kettering, which is HIPAA-compliant and aims to help with diagnosis and admin work. A day earlier, they launched ChatGPT Health for everyday users to analyze test results and prepare for doctor visits, with over 200 million people already asking health questions weekly. Anthropic, led by a former biophysicist, is also focusing on the sector, holding a health care event next week. The push comes as AI labs seek new revenue to offset development costs, but it introduces severe risks of handling sensitive data and potential diagnostic errors in a tightly regulated field.

The revenue play meets life and death

Here’s the thing: this move makes perfect business sense. After saturating the developer tool market, where’s the next big, lucrative, problem-filled industry? Health care. It’s a massive sector drowning in paperwork, complex data, and constant information overload. The potential to streamline admin work for doctors or help patients understand basic info is huge. And for companies like OpenAI and Anthropic, facing astronomical compute costs, embedding their models into hospital systems represents a serious enterprise revenue stream. It’s a logical next step.

But the stakes are just incomparable to helping someone write a Python script or plan a vacation. As the physician consultant in the article starkly put it, a bad travel tip sends you to the wrong train station. A bad medical suggestion could send someone home when they need the ER. That risk of an “AI adverse event,” as the Boston Children’s Hospital exec fears, is what hangs over this entire push. One high-profile failure could trigger a regulatory backlash that sets the whole field back years. They’re playing with fire, and they know it.

The ghosts of health tech past

We can’t ignore the history here. The article quotes a warning that “the battleground of health tech is littered with the bodies of big companies that didn’t know anything about health.” That’s absolutely true. Tech giants have repeatedly marched into medicine with confidence, only to be humbled by the byzantine regulations, entrenched workflows, and sheer complexity of human biology. Health care doesn’t move at Silicon Valley speed. It moves at the speed of clinical trials, peer review, and HIPAA audits.

And let’s talk about the competition. It’s not just OpenAI vs. Anthropic. They’re going up against specialized, legacy tools that doctors already use for transcription or scan analysis. Convincing a busy physician to trust a general-purpose chatbot with a diagnosis, even one with “enhanced citations,” is a monumental task. The AMA report says two-thirds of physicians used AI in 2024, but that likely includes a lot of back-office automation, not direct diagnostic aid. That’s a different beast.

A cautious and watchful rollout

To their credit, the companies seem aware of the pitfalls. OpenAI is carefully positioning its consumer tool, ChatGPT Health, as a “supplement” to professional advice, not a replacement. Their enterprise play is about assisting clinicians, not replacing them. And hospitals like Boston Children’s are conducting studies to compare diagnosis effectiveness with and without AI tools—though they haven’t published results yet. That’s the responsible path.

But the cat is already out of the bag. With 200 million weekly health queries, people are using ChatGPT for medical advice whether it’s designed for it or not. Launching an official health product arguably legitimizes that use, which is a double-edged sword. It allows for better guardrails, but also increases potential liability. It’s a tightrope walk. You can see their health care vision here, and Anthropic’s upcoming thoughts at their event.

The verdict: proceed with extreme caution

So, is this the next big AI market? Probably. The demand is clearly there, both from consumers drowning in medical jargon and doctors drowning in paperwork. The potential benefit for administrative burden alone is worth billions.

But the path to integrating AI into the actual diagnostic chain is fraught with danger. It requires a level of reliability and explainability that today’s generative AI still struggles with. The comparison to self-driving cars in the article is apt: we accept that rare accidents might happen with autonomous vehicles if the overall safety improves. Will we ever accept that from an AI diagnostician? I have serious doubts. The margin for error in medicine feels infinitely smaller to the public. This will be a slow, cautious, and heavily scrutinized battle—not the explosive hype cycle we saw with ChatGPT. The labs have entered a whole new world of risk.