OpenAI's and Anthropic’s new healthcare products highlight potential for access and efficiency but also privacy, reliability, and governance issues

The Buzz This Week

Last week marked a significant escalation in the healthcare artificial intelligence (AI) race as both OpenAI and Anthropic unveiled product releases. The products have both consumers and provider organizations in mind.

OpenAI announced the launch of ChatGPT Health, a new consumer-focused feature, and OpenAI for Healthcare, a set of enterprise AI tools designed for health systems to support clinical and operational workflows. Anthropic launched Claude for Healthcare, an expanded set of AI tools and Health Insurance Portability and Accountability Act (HIPAA)-compliant resources for provider and payer organizations.

According to OpenAI, ChatGPT Health is a new secure feature that allows individual consumers to upload their medical information, connect their wearables, wellness apps, and medical records to get more personalized responses to their health questions. It stops short of diagnoses or treatment recommendations.

Through OpenAI’s recent acquisition of Torch and partnership with b.well, OpenAI aims to create a “unified medical memory” and tackle the elusive personal health record, bringing historically fragmented data sources into a single place. With their existing user base and grounding in secure patient data, OpenAI hopes to solve the problems that doomed personal health record (PHR) efforts in previous decades (e.g., Microsoft HealthVault and Google Health).

At the enterprise level, both OpenAI for Healthcare and Claude for Healthcare are striving to become the AI platform of choice for healthcare organizations. These launches are timely as health systems are building momentum around AI utilization, and physician adoption of AI has nearly doubled within a single year.

OpenAI for Healthcare enables evidence-based AI-support at the point of care. ChatGPT for Healthcare enables the creation of AI-native workflows for both healthcare delivery organizations and health tech partners (e.g., Abridge and Ambience). Supported workflows span clinical and administrative use cases.

Roll-out to a pilot group of health systems has already begun, including Boston Children’s Hospital; Cedars-Sinai Medical Center; Stanford Medicine Children’s Health; AdventHealth; HCA Healthcare; Baylor Scott & White Health; Memorial Sloan Kettering Cancer Center; and UCSF Health. These organizations will be “product testing” various applications of the technology, such as assisting with clinical documentation, streamlining administrative tasks, and supporting patient engagement efforts.

Anthropic has opted to pursue enterprise health offerings exclusively. Anthropic’s Claude for Healthcare, announced in conjunction with this year’s JP Morgan Health Care Conference, builds on the company’s existing AI models with a HIPAA-compliant infrastructure that connects with common medical and scientific databases.

These connections enable Claude to verify coverage requirements, support prior authorization workflows, reduce claim denials, and surface regional coverage differences, according to the company. Anthropic claims that its commitment to building safe, responsible AI makes it well suited to work with health systems, hospitals, payers, and other healthcare organizations in support of their clinical and operational workflows.

Why It Matters

A new wave of healthcare consumerism is arriving as patients show up to appointments armed with their own AI-driven information. Clinicians and health systems are already seeing a shift in the patient-provider relationship. They will need to be prepared with the right tools to ensure that a more informed public leads to better care and experience. The full impact these releases will have on both consumer and health system adoption and utility of AI remains to be seen.

For consumers, ChatGPT Health is appealing because it can provide organized personal health information, tailored answers to health questions, and simplified test results and medical guidance. However, existing services to manage personal health data (such as Apple HealthKit) have had varied success.

Further, while the personal trust environment has evolved and comfort levels with technology and data sharing have grown, concerns about security and privacy remain. Data and information shared with AI chatbots and other technologies are not within the scope of HIPAA and therefore not currently protected.

The unreliability of large language model-based agents applied to healthcare is another primary concern. A recent study found that models are designed to prioritize being helpful over medical accuracy. Possibly misleading or incorrect information in clinical guidance or interpretation of clinical results are real risks for both patients and providers.

Anthropic, OpenAI, and other leading AI companies caution that their systems can make mistakes and should not be substitutes for professional judgment. Character.AI and Google recently agreed to settle a lawsuit alleging their AI tools contributed to worsening mental health among teenagers who died by suicide.

Health systems and health plans should view AI integration as a core pillar of the organization’s strategy, reflective of its strategic and operational priorities, rather than a standalone IT project or pure software implementation. One-off projects that add AI in a single department or replace a single vendor with a new AI tool or workflow can be counterproductive. Instead, an end-to-end rethinking of people, processes, and technology with an AI-native approach is required.

For many organizations, this will mean building workflows from the ground up, removing outdated processes, shifting operating models, and prioritizing change management while pursuing new AI use cases.

Health systems also will need to establish their own approach to governance and observability (i.e., monitoring, measurement, guardrails) as they implement new programs. As the industry moves into a newly connected environment, a multi-pronged approach will be critical to address technical and ethical principles and establish a technical infrastructure with appropriate guardrails built in.

Keeping up with the dizzying speed of digital and AI transformation is a tall order. By selecting a few initial use cases and trusted partners to collaborate with, organizations can sustain focus and momentum. Maintaining pace while staying true to the organization’s mission and principles will be critical for health system and health plan leaders in the coming months.