ChatGPT Health and Taiwan take different paths to health data

OpenAI began rolling out Health in ChatGPT to US users aged 18 and older on July 23, 2026, allowing them to connect Apple Health and supported medical records. Taiwan uses a different route for records held through its national insurance system: people authorize selected data to move from My Health Bank to an app reviewed by the government.

The two services are not direct substitutes. Health in ChatGPT is a consumer AI service whose operator manages the connection and its partners, while Taiwan's software development kit (SDK) is a government-controlled gateway used by multiple app operators. Their differences are clearest in who approves access, where data go and which organization can act when a connection is misused.

ChatGPT connects records inside one consumer service

OpenAI's July launch notice says US adults can connect Apple Health and supported records from US hospital systems, One Medical or Function Health. ChatGPT asks for permission by default before using connected information in a response, although an account holder can grant continuing access. The information can inform conversations outside the Health area after permission is given, reversing the early product design in which the dedicated space was the only place that could use it.

OpenAI says connected medical records, Apple Health information and conversations that use those data are not used to train its foundation models or target advertisements. Disconnecting a source stops future access and starts deletion of synchronized data from OpenAI's systems within 30 days. Information already placed in conversation history remains until those conversations are deleted.

Taiwan places a gateway between records and apps

Taiwan's National Health Insurance Administration (NHIA), which runs the territory's single-payer health insurance program, built My Health Bank to give insured people access to their claims-linked health records. The agency says its SDK lets a person select records covering visits, medications and test results for a specified period and authorize their transfer to a trusted third-party app.

The app, rather than the NHIA app, starts the connection. The person then verifies identity, chooses the categories and period to share, and grants a one-time authorization. This model does not put every participating app inside one government product; it controls the point at which an outside operator receives data from My Health Bank.

The NHIA's current service page says new first- and second-stage applications to connect through the SDK are temporarily suspended because of operational adjustments. The notice does not give a resumption date. Existing connected apps remain listed separately.

Review and consent do different jobs

OpenAI controls which record sources and apps its service supports. Its published privacy notice describes how Health Features collect and use linked records, but it also makes clear that processing is not confined to a single corporate system. OpenAI says authorized personnel and trusted service providers may access Health Feature data for model-safety work unless the user opts out, and vendors may process data for functions such as hosting, support and safety monitoring.

Taiwan's SDK adds a public gate before an app can connect. The NHIA says connected apps must pass its review and meet Taiwan's published mobile-app security requirements; transfers use encryption, each release requires authorization, and the agency retains access logs for investigations. The agency may terminate an app's SDK service if medical information is leaked or misused.

Those controls address access and traceability, not the clinical quality of an app's interpretation. An app can satisfy identity, encryption and recordkeeping requirements without proving that every AI-generated explanation is accurate. OpenAI likewise reports product testing with physicians, but its July notice says ChatGPT can still make mistakes and does not replace professional judgment.

Accountability follows the data path

Health in ChatGPT gives one account holder a set of controls managed through OpenAI: permission prompts, disconnection, memory settings and deletion. Its responsibility chain also includes medical-record sources and service providers described in OpenAI's privacy notice. Which consumer-health privacy rules apply can depend on where a user lives, so the US launch does not establish a single complaint or enforcement route for every future market.

The Taiwan model divides responsibility between the NHIA and each connected app. The agency can review access, retain transfer logs and end the SDK connection. Once a third-party app has received authorized records, however, a person seeking to stop processing or request deletion must contact that app's operator, according to the NHIA.

The current app count is less certain than the governance design. The source report cited 31 organizations and 64 apps in January 2026, but the NHIA's current public directory names connected apps without stating an overall total. APPI News could not reconcile the earlier figure with the live directory at the time of writing.

Two models, two limits

OpenAI's model can make linked information available within a broad AI service under one set of account controls. Taiwan's model gives a public agency control over entry to a national record gateway while leaving services and downstream storage to separate apps. Neither structure alone establishes the accuracy of an AI-generated health explanation.

The comparison also has a geographic limit. Health in ChatGPT currently supports connected medical records in the United States, while My Health Bank is tied to Taiwan's insurance system. Claims about which model offers stronger accountability therefore depend on the specific operator, data recipient and law involved, not only on whether the gateway is run by a company or a government agency.