When people think about AI in medicine, a few images may come to mind: a patient at home trying to figure out a rash, an algorithm reading a CT scan, or a doctor using an AI scribe to draft a note.

Anthropic, the San Francisco company behind Claude AI, has spent much of 2026 building something different. It is focusing on healthcare’s plumbing: coverage rules, billing codes, medical records and research pipelines that run the American healthcare system. It is also betting that its models can speed up drug discovery.

Over the past year, the company has launched a healthcare product line, let consumers connect their medical records to Claude, released a pharmaceutical research workbench, and opened its own biology lab. In the past two weeks alone, it partnered to give free AI-powered clinical search to doctors in roughly 100 countries and joined a federal clinical AI moonshot.

Here’s what’s happening, and what it means for doctors and patients.

Claude Can Now Help With Billing, Coverage And Prior Authorization

In January 2026, Anthropic unveiled Claude for Healthcare. The launch came days after OpenAI introduced ChatGPT Health.

The centerpiece was a set of connectors to databases that govern coverage and billing. For example, a connector to the CMS Coverage Database lets Claude look up local and national coverage determinations, support prior authorization checks and help build claims appeals. Others link Claude to ICD-10 diagnosis and procedure codes and to the National Provider Identifier Registry.

Hospitals, insurers and health tech companies can use these tools for work involving protected health information (PHI) through HIPAA-ready enterprise products. A few customers Anthropic has highlighted: Banner Health, Genmab, Heidi Health and Novo Nordisk.

The company said its tools could speed up prior authorization requests “so…patients can get life-saving care more quickly.” Claude pulls coverage requirements, checks clinical criteria against the patient’s record and then proposes a determination with supporting materials for a human reviewer at the payer.

Given where healthcare is today, this focus makes sense. Administrative burden is one of the costliest and most frustrating problems in American healthcare. And prior authorization sits near the top of most doctor’s list of complaints.

The tools are being offered to both sides. Insurers can use AI to review requests faster. Hospitals and practices, by contrast, can use it to build stronger appeals.

Whether that shortens delays for patients or simply accelerates an automated arms race between payers and providers is an open question.

Patients Can Now Connect Their Medical Records To Claude

Alongside administrative tools, U.S. subscribers on Claude’s paid Pro and Max plans can connect lab results and health records through partners such as HealthEx and Function, as well as Apple Health and Android Health Connect. Claude can summarize a person’s medical history, explain test results in plain language, spot trends in the data and help prepare questions for appointments.

Anthropic says these integrations are opt-in. That means that users choose exactly what to share and can disconnect at any time. They also say they don’t use the users’ health data to train models.

Many patients are already doing a cruder version: pasting screenshots of lab portals into chatbots at all hours of the day and night. I’ve written about how patients can do that more safely .

Still, there are real tensions with consumer-based AI assessments. Anthropic’s own usage policy says that in high-risk uses such as medical diagnosis or patient care, a qualified professional should review AI output. Claude, according to Anthropic, is designed to include disclaimers, acknowledge uncertainty and direct people to clinicians. Yet a patient asking about their own lab results at midnight might not have the luxury of a clinician in the loop.

Additionally, when a patient chooses to send their own records to a consumer app, the data generally falls outside the HIPAA protections that govern hospitals, insurers and their vendors. What protects it instead is the company’s own privacy policy--which can change--and a patchwork of state consumer health privacy laws.

Finally, a central, unresolved issue is accountability when a plain-language AI explanation turns out to be wrong.

Anthropic Is Betting Big On Drug Discovery

Anthropic launched Claude for Life Sciences in October 2025. In January, it pushed into clinical trial operations, adding connectors to Medidata, ClinicalTrials.gov and Open Targets, along with a tool for drafting trial protocols that accounts for FDA guidelines.

In June, it went further with Claude Science: a workbench that brings scientific databases, lab tools and computing power into one place for pharmaceutical researchers. At the launch event, Novartis CEO Vas Narasimhan, who joined Anthropic’s board in April, estimated that platforms like it could shorten a drug’s path to approval from about 12 years to seven or eight and double the industry’s success rate from roughly 8% to 16%.

In September, Anthropic also announced a new life sciences research group with its own lab in the Bay Area. In early results, the company said Claude agents searching a massive DNA sequence database flagged a previously uncharacterized enzyme system with repeat patterns reminiscent of CRISPR. Yet Anthropic has been candid about the limits: “we don’t yet know its function.” To date, findings on this new system have only been released as a preprint, not a peer-reviewed paper.

Separately, the company launched a Life Sciences Verification Program. The goal is to vet research teams’ credentials, security and ethical oversight and then give approved teams access to models with fewer safeguards on biology-related work that Anthropic’s generally available models block.

From my perspective as a physician, claims around AI drug discovery should be met with caution. Drug development can fail at any stage, and failure rates are especially high once a drug is tested in humans.

Bristol Myers Squibb CEO Chris Boerner agrees and has warned the industry not to “get over our skis.” Whether AI can meaningfully speed drug development very much remains to be seen.

Global Health And Federal Partnerships Are Expanding

Recent moves at the company have extended beyond the U.S. In September, Anthropic and OpenEvidence announced a free, locally adapted version of OpenEvidence’s clinical search tool. The target is clinicians in about 100 low- and middle-income countries , including Uganda, Haiti and Mongolia, where access to medical literature and specialists is thin. Anthropic is supplying the underlying models, along with Claude credits, as well as dedicated engineers and funding. OpenEvidence adapts the tool to local disease patterns and resources.

Anthropic president Daniela Amodei told Reuters that market incentives alone “would not let it happen.” Yet critics have noted that tools built largely on evidence from wealthy countries may not reflect practice where tests and drugs are scarce.

Anthropic has also described how the World Health Organization’s Africa office, the vaccine coalition CEPI and the Democratic Republic of Congo’s national biomedical research institute are using Claude in the response to the Bundibugyo Ebola outbreak. According to the company, one WHO tool cut preparation of a daily situation report from a full day to under an hour.

Back home in the U.S., Anthropic has joined a clinical AI moonshot led by ARPA-H, the federal agency that funds high-risk, high-reward health research. The program, called ADVOCATE, is a four-year, $62.7 million effort to build the first FDA-authorized AI agents that can support heart failure patients between visits. This includes adjusting medications and escalating to the care team when needed.

All of this is happening as Anthropic barrels toward the public markets. The Wall Street Journal reported in August that Anthropic was telling prospective investors that its health and biology work would improve public sentiment toward AI.

Yet its IPO prospectus also warns that advanced AI could pose “catastrophic or existential risks.” STAT has highlighted the tension between a company that warns AI could one day lead to catastrophe and its aggressive move into a sector where AI regulation remains in flux.

Here’s What Doctors And Patients Should Watch

In my view, three things will determine whether Anthropic’s push into healthcare actually moves the needle.

The first is evidence. Benchmarks and demos are not enough. Outcomes and error rates from real deployments are needed to show proof of value. Second, with respect to prior authorization, it matters whether AI on both sides reduces denials and delays or simply accelerates the fight between payers and providers. Third is consumer safeguards. Patients need clear disclosure of where their health data goes and what happens when a model gives the wrong answer.

For clinicians evaluating any AI product, the questions should be which model it runs on, how it was validated and who signs the business associate agreement that legally obligates the vendor to protect patient data.

Patients connecting records to Claude, or any AI for that matter, should treat its explanations as a starting point for the visit, not a substitute for it.

Ultimately, I think Anthropic has picked the right problems at the right time given the strengths of the platform. Whether it solves them is a question that only time (and results) can answer.