18 July 2024, Brandenburg, Cottbus: A live interaction between a simulated patient (Doris Härtel) and a robot can be seen at a press event at the Carl-Thiem-Klinikum Cottbus. (Photo by Patrick Pleul/picture alliance via Getty Images)
dpa/picture alliance via Getty Images
What if AI could speak the language of medical billing? Is the healthcare industry ready for AI assistants that can understand the byzantine world of medical coding, insurance approvals and patient records?
Anthropic’s vote is yes. This weel, the San Francisco-based company announced Claude for Healthcare, launching a suite of AI tools for doctors, insurers and patients navigating the maze of America’s notoriously complex medical system.
The timing is noteworthy: the rollout comes just one week after rival OpenAI introduced its own healthcare product, hinting at an intense race among AI developers to crack one of the economy’s most regulated — and lucrative — sectors.
Does Healthcare Need More Than Generic AI?
While general-purpose chatbots often confidently hallucinate medical advice, a more encouraging sign is that Claude for Healthcare connects directly to verified American medical infrastructure.
The system plugs into databases including the Centers for Medicare & Medicaid Services Coverage Database — which determines what procedures insurance will pay for — as well as ICD-10 medical coding standards, the alphanumeric system doctors use to bill for everything from a sprained ankle to open-heart surgery, and PubMed’s library of biomedical research papers.
Theoretically, that means AI could help draft prior authorization requests for healthcare providers — i.e. the paperwork insurance companies require before approving certain treatments or medications. According to a 2024 American Medical Association survey, prior authorizations consume a whopping average of 13 hours of physician and staff time per week, delaying patient care and contributing to clinician burnout. If AI can streamline even a fraction of that administrative burden, the productivity gains could be substantial.
“These tools can be used to speed up prior authorization requests so that patients can get life-saving care more quickly … and help with regulatory submissions so that more life saving drugs can come to market faster,” the company said.
What’s Actually New Here?
Anthropic’s healthcare push builds on Claude Opus 4.5, its flagship AI model released late last year.
According to Anthropic’s internal testing, this version shows marked improvement on simulated medical and scientific tasks compared to earlier iterations with fewer factual errors.
The product includes what Anthropic calls Agent Skills — pre-built tools that developers can customize for specific healthcare workflows. One example automates parts of the prior authorization process, while another helps programmers build applications using Fast Healthcare Interoperability Resources, the emerging standard for exchanging medical data between different hospital systems, insurance platforms and electronic health records.
Anthropic is introducing integrations that let U.S. subscribers on its Pro and Max subscription tiers connect Claude to their personal health records. The firm has partnered with HealthEx and Function Health and is launching beta integrations with Apple HealthKit and Android Health Connect through Claude’s mobile apps. The AI firm said that health data accessed through these connections isn’t stored in Claude’s memory banks or used to train future AI models.
“These integrations are private by design,” the company stated. “Users can choose exactly the information they share with Claude, must explicitly opt-in to enable access and can disconnect or edit Claude’s permissions at any time.”
Notably, all of this infrastructure is also HIPAA-ready, meaning it meets the Health Insurance Portability and Accountability Act’s requirements for protecting patient privacy — a compliance framework that is table stakes for any tech company serious about operating in healthcare.
AI Healthcare Land Grab
OpenAI’s recent healthcare announcement, coupled with billion-dollar valuations for AI-focused health startups like Abridge and Sword Health, indicates belief that this market is approaching an inflection point.
U.S. healthcare spending was estimated at $4.9 trillion in 2023, or a staggering $14,570 per person, according to the Centers for Medicare & Medicaid Services. AI tools that demonstrably reduce paperwork and speed up insurance approvals, or help patients better understand their conditions, could potentially improve outcomes by a sizeable measure.
On the other hand, healthcare data is fragmented across incompatible systems, liability concerns loom large and physicians have legitimate worries about AI errors in high-stakes clinical decisions. AI tools that promised to revolutionize radiology and diagnostics have seen discouraging real-world results, often amplifying existing biases in medical data or failing to integrate smoothly into clinical workflows.
Patients & Providers
The near-term promise for patients seems more simple: asking plain-language questions about lab results, medication side effects or treatment options with AI that can reference actual health history rather than offering generic WebMD-style advice. But meaningful value will depend on execution, real outcomes and whether people actually trust AI with their most personal medical information.
Meanwhile, the calculus for healthcare organizations involves cost savings versus implementation headaches: Hospital systems and insurers are already experimenting with AI for administrative tasks, but sustainable adoption requires these tools to work reliably within existing workflows rather than creating new problems and dangers.
Anthropic is also expanding Claude’s life sciences capabilities with connections to platforms including Medidata, which is used in clinical trials, ClinicalTrials.gov, which is the federal database of medical studies and bioRxiv, where researchers share preliminary findings. New agent skills could help draft FDA- and NIH-compliant clinical trial protocols or monitor study performance, which are currently tasks that require specialized expertise and considerable time.
The healthcare AI market is still nascent with more promise than substantive, proven results. As major AI companies commit serious resources to medical applications, the sustainability of this trend could depend on whether the tools prove accurate and reliable enough for high-stakes medical workflows — a bar that, in the past, has proven difficult to clear.