ChatGPT

  • |

    ChatGPT Wants Your Medical Records, But Not the Responsibility

    ByLee Keybum

    Lee Keybum reads the terms so you do not have to, and OpenAI’s new Health feature comes with a very modern trust bargain: ChatGPT can become familiar with your connected medical records, lab results, prescriptions, Apple Health data, and health conversations, but it still wants to be treated as support rather than medical care. OpenAI announced the U.S. rollout on July 23, describing Health in ChatGPT as a place for personalized guidance and health-related conversations. That is a lot of intimate information to hand to a general-purpose chatbot before it politely reminds you that it is not intended to diagnose or treat anything.

    The contradiction is not hidden; it is laminated into the product. The more information ChatGPT can use, the more naturally it can sound like the friend who knows why every pill is in the cabinet and which lab result made you stare at the ceiling. But OpenAI says the service does not replace qualified medical professionals. In ordinary-person language, the platform wants the context required to sound medically informed while preserving an escape hatch if the conversation wanders into actual medical responsibility.

    OpenAI’s health privacy policy also says connected health data is not used to train its foundation models or target ads by default. That matters, and “by default” matters too. It is not a promise that every possible privacy concern has evaporated into the cloud. It means the company’s stated bargain is narrower: users may authorize sensitive data connections for the feature, while OpenAI says those connections are treated differently from ordinary ChatGPT data for model training and advertising.

    For users, the practical question is not whether the chatbot can produce a soothing paragraph about wellness. It is whether convenience quietly turns the app into the most informed entity in a person’s medical life without giving that entity the accountability people normally expect from medical care. A doctor has credentials, professional rules, and a human being attached to the decision. ChatGPT has a privacy policy, a disclaimer, and a remarkable talent for making a sentence sound settled before the facts are.

    That is the subscription-barnacle version of artificial intelligence: first it asks for your bloodwork, sleep history, medication list, and trust; then, if the answer is wrong, it becomes a very confident autocomplete with no medical license. OpenAI may call Health a support tool, but the user experience is built to feel personal and informed. The company wants the data that makes the system sound like a doctor while keeping the responsibility label safely out of frame.

  • Your Doc’s Been Promoted to “Elevated Errors”: The ChatGPT Upload/Download Rollback That Ate Tuesday

    ByLee Keybum

    I didn’t ask ChatGPT to join a software improv troupe. I asked it to do the one job that matters in my week: upload the file, get the task done, and download the result like a normal human being with places to be.

    Instead, my document got the upgrade nobody wants: elevated errors. And the fun part (for the platform) is that this isn’t some obscure corner case. It’s the core file-moving step—uploading and downloading—where the status page publicly waved the “it’s basically fine” flag while the actual upload/download experience turned into a waiting-room performance.

    That’s the contradiction in plain English: the platform narrative is “everything’s working,” but the workflow reality is “your doc is now the guest of honor at the spinner buffet.” On Tue, Jun 23, 2026, OpenAI’s status communication flagged “elevated errors” for ChatGPT uploading/downloading files, and then later marked the incident as resolved/fully recovered—after my time had already been reassigned to staring at progress bars like they’re going to apologize.

    So sure, the status page says it’s resolved. Great. That means the problem finished having a problem. The only thing that reliably showed up on schedule was my subscription clock—while my deliverable was stuck in the AI pipeline doing the exact opposite of “promoted to done.” And if you’re wondering why this feels like a rollback: congrats, your Tuesday got downgraded to file-transfer archaeology.

End of content

End of content