Anthropic Chases OpenAI Into Healthcare. The Graveyard Is Crowded.

The timing was not subtle. OpenAI announced ChatGPT for Health on a Monday. Five days later, Anthropic fired back with Claude for Healthcare. The setting for both: JP Morgan's annual healthcare conference in San Francisco, where hospital executives hunt for capital and AI companies hunt for customers. HIPAA-ready infrastructure. Clinicians, insurers, patients. The pitch from both companies could have been written by the same marketing team.

And both entered a market littered with the remains of previous attempts by horizontal technology companies to fix American healthcare.

Anthropic is currently negotiating a $10 billion funding round that would value the company at $350 billion. The healthcare push is not charity work. It is a bet that Claude can succeed where Microsoft HealthVault and Google Health failed, where countless electronic health record integrations have stalled, where the promise of consolidated patient data has been made and broken for two decades.

The pitch sounds familiar. Connect scattered medical records. Let patients ask questions in plain language. Help clinicians find information faster. Reduce prior authorization friction. But Anthropic is making one argument that its predecessors could not: the models are now smart enough to do the hard work of translation and integration that human engineers kept failing at.

"Maybe we were all a bit misguided in trying to put all of the health records in one place," Eric Kauderer-Abrams, Anthropic's head of biology and life sciences, told Fierce Healthcare. "That's too hard. Make the models go and do the hard work of figuring out how to interface and communicate and make sure all the connections are in place."

That is either an insight or wishful thinking. The answer matters a great deal.

The Safety Card

Anthropic has been selling itself as the responsible AI company since 2021, when a group of OpenAI researchers walked out and started their own lab. Dario Amodei runs the place. He trained as a biophysicist, which matters here. The company has invested heavily in "Constitutional AI" methods designed to reduce harmful outputs. In a market where OpenAI moves fast and breaks things, Anthropic moves deliberately and publishes safety research.

Healthcare is where that positioning pays dividends. Or is supposed to.

"Anthropic is a very natural fit for the healthcare and life sciences world because our identity as an AI company is built around safety and responsibility and rigor and reproducibility," Kauderer-Abrams said. These are words calculated to resonate with hospital compliance officers and pharmaceutical regulatory teams. They are also words that create expectations.

Banner Health, one of the largest nonprofit health systems in the United States, has 22,000 clinical providers using Claude. Anthropic claims 85% of them report working faster with higher accuracy. Stanford Healthcare signed on. The pharma giants followed: Novo Nordisk, Sanofi, AbbVie, Genmab. The enterprise traction looks real enough on a slide deck.

But the consumer side is where the safety rhetoric gets tested. Claude Pro and Max subscribers can now connect their health records through HealthEx, a startup that aggregates data from more than 50,000 care provider organizations. Apple Health and Android Health Connect integrations are rolling out in beta. Patients can ask Claude to interpret lab results, summarize their medical history, or prepare questions for doctor visits.

The promise sounds clean. The reality of healthcare data is not. Somewhere in America right now, a nurse is clicking through six windows to find a potassium level. A fax machine is humming in a back office, transmitting referral documents in 2026 because the two hospital systems never figured out how to share records electronically. A patient is staring at a portal that shows half their medical history, the other half trapped in a system that closed five years ago. This is the environment Claude is entering.

Anthropic says it will not train its models on healthcare user data. The company says data accessed through these integrations is not stored in Claude's memory. These are reassuring claims. They are also claims that will be verified only in retrospect, when something goes wrong or an audit reveals otherwise.

Mike Reagin, Banner Health's chief technology officer, stated the obvious: "Safety is non-negotiable in healthcare." But safety in healthcare is also contextual. A model that hallucinates a medication interaction that never existed is unsafe. A model that fails to flag a dangerous drug combination because it lacks context is also unsafe. The gap between "fewer hallucinations" and "safe enough for medical decisions" is large and poorly mapped.

The Graveyard

Microsoft launched HealthVault in 2007 with the promise of letting patients store and share their medical records in one place. Twelve years later, the company pulled the plug. Google Health tried something similar in 2008. Four years. That is how long Google lasted before shutting it down. They came back later with a hospital partnership—Ascension, one of the largest Catholic health systems—and promptly drew regulatory scrutiny for how patient data moved between the two organizations. The obituaries write themselves: horizontal technology companies enter healthcare with grand visions, get bloodied by the regulatory complexity, and retreat.

Anthropic is making a different bet. The previous failures, Kauderer-Abrams argues, stemmed from trying to centralize records. That was the wrong approach. Google and Microsoft tried to build a new Library of Alexandria for health data, moving everyone's records into one grand repository. Anthropic is not building a library. It is building a librarian who can run between the stacks, fetching what you need from wherever it happens to live.

HealthEx, the first consumer health record integration for Claude, operates on this model. Users connect their patient portal logins. HealthEx unifies records across providers. When users ask Claude health-related questions, the model uses Anthropic's Model Context Protocol to retrieve only the relevant portions of the record for each specific query. Allergies for a medication question. Recent labs for a cholesterol inquiry. Doctor notes for a symptom pattern.

"We're giving every American a safe, private way for them to use their health data with AI," Priyanka Agarwal, HealthEx's CEO and a physician, told Fortune. The company is small. The promise is enormous. And the gap between a startup partnership and a scaled healthcare solution is where previous attempts have died.

The fundamental problem has not changed. Healthcare data in America is fragmented by design. Providers run different electronic health record systems, and those systems talk to each other the way rival gangs talk to each other—grudgingly, through intermediaries, with a lot of dropped messages. FHIR, the supposed standard for exchanging healthcare data, works when it works. Patients trying to access their own records often find half the story missing, the other half buried in PDF scans of faxed documents.

Anthropic is betting that Claude is smart enough to translate between systems, surface relevant information, and present it in ways patients and clinicians can use. The connector list reads well: CMS coverage databases, ICD-10 codes, the National Provider Identifier Registry, PubMed's 35 million papers. On a spec sheet, it looks like infrastructure. In practice, connectors are just pipes. Whether anything useful flows through them is a different question.

What's Actually in the Box

Claude for Healthcare is not a new model. It is not even a new product. What Anthropic announced is a set of database connectors and enterprise features wrapped in HIPAA-ready infrastructure.

"These are not new models or even new products," Kauderer-Abrams acknowledged. What Anthropic announced is connectors—pipes between Claude and the databases clinicians already use. That is the pitch. Credit where due: most AI companies would have dressed this up as a revolution. Anthropic called it what it is. Anthropic is not claiming to have built something novel for healthcare. The company is claiming that Claude Opus 4.5 is now good enough, and the surrounding infrastructure is now complete enough, that healthcare organizations can deploy it safely. The bet is on model quality and integration polish, not on healthcare-specific innovation.

For enterprise customers, the value proposition is administrative automation. Prior authorization requests that currently take hours of staff time could be accelerated if Claude can pull coverage requirements from CMS databases, match them against patient records, and generate the appropriate documentation. Claims appeals could be drafted faster. Clinical trial protocols could move from days of drafting to about an hour, according to Anthropic's demonstration.

For life sciences companies, Claude now connects to Medidata for clinical trial data, ClinicalTrials.gov for trial registries, bioRxiv and medRxiv for preprints, and Open Targets and ChEMBL for drug discovery databases. The company is positioning Claude as a research collaborator that can work across the full drug development pipeline, from target identification through regulatory submission.

The enterprise customers are real. The use cases are plausible. But "plausible" is doing significant work in that sentence. Healthcare is full of plausible AI use cases that never achieved sustainable adoption. The question is whether Anthropic can convert pilots into production deployments that justify the $350 billion valuation investors are considering.

The Horizontal Problem

Brendan Keeler, who writes the Health API Guy newsletter, described OpenAI's healthcare announcement as "putting out the shingle." The phrase applies equally to Anthropic. Both companies are signaling that they are open for healthcare business. Neither has yet built the deep vertical integration that would make them essential to clinical workflows.

Horizontal AI companies face a structural challenge in healthcare. They optimize for general capability. Healthcare rewards specificity. A model that can write poetry and debug code and summarize research papers and generate marketing copy is impressive. A model that reliably distinguishes between concerning and benign lab value patterns, that understands the fine print of drug formularies across different insurance plans, that can navigate the politics of a hospital system's documentation requirements—that is a different kind of achievement.

Anthropic's answer is connectors. Plug Claude into the databases clinicians use. Let the model's general intelligence work on domain-specific problems. The logic is seductive: skip the domain expertise, trust the general reasoning. It is also the exact logic that gets horizontal companies killed in vertical markets.

But the incumbents are not idle. Epic controls electronic health records at most major hospital systems, and the company has been quietly building its own AI features for years. The specialized startups are flush too—Abridge and Sword Health both crossed into multibillion-dollar territory. These players know the workflows cold. They have spent years learning exactly where software breaks when it meets a busy nurse at 3 AM.

Anthropic brings something they lack: a foundation model that improves with every general capability advance the company makes. When Claude gets better at reasoning, healthcare applications get better too. That edge is real.

What Anthropic does not have is patience. The company is racing OpenAI for healthcare market share while simultaneously negotiating a funding round that implies extraordinary growth expectations. The pressure to show traction will be intense. Healthcare does not move at venture capital speed.

Banner Health's 22,000 clinical providers represent a significant deployment. But Banner is one health system. The American healthcare market contains thousands. Convincing cautious healthcare executives to bet on AI from a five-year-old company—even one valued at $350 billion—will require sustained effort over years.

Anthropic has the model. The company has the connectors. What remains unclear is whether it has the organizational stamina for a market that has broken better-funded and more patient companies than itself. The hospital administrator who approves a Claude deployment in 2026 will want to know the company will still exist, and still support the product, in 2030 and beyond.

The next six months will reveal how deep the healthcare commitment runs. Connectors are a start. Sustained integration is the test.

Marcus Schuler

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai