OpenAI and Anthropic staked claims in the lucrative healthcare market, each company playing to its strengths by targeting different audiences.
What’s new: OpenAI presented ChatGPT Health, a consumer-focused version of its chatbot that can retrieve a user’s medical information, with upgraded data security and privacy protections. A few days later, Anthropic unveiled Claude for Healthcare, a set of tools largely designed to help healthcare professionals find medical reference information in databases and speed up paperwork.
ChatGPT Health: OpenAI’s offering is a health and wellness chatbot that builds on the updated OpenAI for Healthcare API. It’s designed to help consumers understand their own healthcare information including medical tests and doctors’ instructions as well as data from patients’ own wearable devices. OpenAI says it built it over two years based on feedback and advice from 260 physicians in 60 countries.
- Architecture: ChatGPT Health is a sandbox inside ChatGPT with its own memory, connected apps and files, and conversations. It can use data from ChatGPT conversations outside the sandbox, but not vice versa. OpenAI did not specify the models used or whether they were fine-tuned on health data or use system prompts specific to health conversations.
- Functionality: The chatbot explains lab results, prepares questions to ask physicians, interprets data from wearable devices, and summarizes care instructions. Users can share their medical information with the system, which stores it as context. They can do likewise with health and wellness data from Apple Health, Function, MyFitnessPal, and other platforms.
- Privacy: The system added extra security, isolating and specially encrypting sensitive data. OpenAI partner b.well securely fetches personal data like test results, medications, and medical history from doctors and hospitals. Health conversations and context are not used to train OpenAI models.
- Availability: ChatGPT Health is available via waitlist to free and paid users outside the European Union, Switzerland, and United Kingdom. OpenAI said it would be available for all desktop and iOS users within weeks.
Claude for Healthcare: Anthropic’s entry is designed for providers. It relies on two elements of the Claude platform: Connectors give Claude access to third-party platforms (in this case, medical databases), and agent skills help Claude perform specific tasks (in this case, sharing medical data and producing medical documents). Anthropic is testing a feature that allows some users to connect their medical information.
- Data access: Claude for Healthcare connects to the following databases: CMS Coverage Database, which manages healthcare claims for members of U.S. public-health plans; ICD-10, a reference manual of codes for diagnoses and procedures; and the National Provider Identifier Registry, which verifies physicians and other healthcare providers. At a user’s discretion, it can read patient lab results and health records using HealthEx and Function protocols. Users can also connect to wearable data from Apple Health and Android Health Connect.
- Functionality: Two skills, FHIR development and prior authorizations, improve paperwork management. FHIR is a specification for electronic exchange of healthcare records and other data. Prior authorizations are required by insurance companies to reimburse some prescriptions and procedures. Healthcare professionals can use these skills to get insurers to approve prescriptions faster, appeal denied insurance claims, read and write patient messages, and reduce administrative overhead.
- Privacy: The new skills and connectors comply with Health Insurance Portability and Accountability Act (HIPAA), the U.S. law that governs medical privacy and data security.
- Availability: Connectors and skills for healthcare practitioners are available to all Claude subscribers, while connectors to patient information are limited to paid subscribers in the U.S.
Behind the news: Many companies have brought to market AI systems for physicians and lab technicians, from voice assistants for doctors to vision models that detect cancers. But consumer-facing approaches have encountered snags. Recently, Google pulled AI summaries that were found to provide incorrect health information. Some U.S. states have sought to regulate chatbots that provide medical advice.
Why it matters: Healthcare is a large potential market for AI. In industrialized nations, healthcare consumes over 10 percent of the gross domestic product, and many countries are facing shortages of medical staff, aging populations, and bureaucratic tangles. In the U.S. alone, the healthcare industry employs 17 million people and accounts for spending of around $5 trillion annually, including $1 trillion in administrative costs. OpenAI’s focus on serving patients and Anthropic’s on healthcare professionals accord with the companies’ relative strengths in the consumer and enterprise markets respectively.
We’re thinking: Europe’s General Data Protection Regulation (GDPR), which imposes strict limits on how medical organizations can share patient data with third parties, remains a hurdle neither company has yet approached. While GDPR protects privacy to a modest degree, it is slowing down the rate at which EU citizens gain access to AI innovations.