Paubox flags OpenAI privacy change

Healthcare AI governance got a little more concrete this week. Not because Congress passed a new law, but because OpenAI quietly changed its U.S. privacy policy on April 30 and Paubox called out what that means for anyone building around patient data. The issue is not that OpenAI is suddenly selling chats to advertisers. It says it is not. The issue is that ads, health workflows, and privacy promises now live much closer together inside the same product family. (paubox.com) ### What changed in OpenAI’s policy? The new language says OpenAI may receive information from advertisers and other data partners to measure and improve ads shown to Free and Go users, including purchase information from those advertisers. That is a meaningful shift because it moves the company from “we show ads” into “we also measure what happens after them.” Op(paubox.com) now more explicitly two-way on the measurement side. (paubox.com) ### Who actually sees ads? Right now, the ad test is for logged-in adult users on the Free and Go tiers. Plus, Pro, Business, Enterprise, and Education tiers are excluded. OpenAI also says users can avoid ads by upgrading, or in the Free tier by opting out in exchange for fewer daily free messages. That matters because the privacy posture is no longer uniform across ChatGPT. It now depends in part on which tier a person uses. (openai.com) ### Isn’t health data walled off? Mostly, that is the point of ChatGPT Health. OpenAI launched it on January 7 as a separate health-focused experience with added protections, purpose-built encryption, isolation, and a promise that Health conversations are not used to train foundation models. Users can connect medical records and wellness apps, and OpenAI says Health operates as its own compartment. If a regula(openai.com)t moving into Health for the extra protections. (openai.com) ### So why are healthcare people still uneasy? Because governance problems rarely come from the clean architecture diagram. They come from human behavior. A clinician, staff member, or patient does not always know which tier they are on, which workspace they are using, or whether a conversation belongs in a protected health flow versus a general-purpose chat. Paubox’s point is basically that the product bo(openai.com)h is hidden, people will take the convenient one. (paubox.com) ### Are ads showing next to health chats? OpenAI’s ad policies say sensitive conversations, emotionally reliant contexts, and mental and personal health conversations are ineligible for ads. It also says healthcare ads and health claims are disallowed during the launch phase. But there is an important nuance — the policy was updated on April 29 to say medical, leg(paubox.com) are. That is more precise, but also more complicated. (openai.com) ### Why does that nuance matter so much? Because “health” is not one thing. A conversation about step counts or diet logging is different from a cancer scare, but both can happen in the same interface. OpenAI is trying to separate regulated, vulnerable, and merely relevant contexts more finely. That may improve ad targeting and product flexibility. The catch is that healthcare organizations usually want the opposite i(openai.com)biguity around sensitive use. (openai.com) ### What does this mean for healthcare builders? It means privacy can’t just be a policy page anymore. It has to show up in product design — visible boundaries, clear disclosures, separate environments, retention controls, and staff rules about where protected health information can and cannot go. Paubox ties that to a wider compliance backdrop, including federal transparency and discrimination-risk rules plus newer s(openai.com)w processes. (paubox.com) ### Bottom line? The story is not “OpenAI ads are in your medical chart.” That is not what happened. The real story is that OpenAI now runs a health-specific experience with stronger protections while also expanding ad infrastructure elsewhere in ChatGPT. For healthcare teams, that makes governance a product question, not just a legal one. (openai.com)