OpenAI unveiled ChatGPT Health on January 7, 2026, pitching it as a HIPAA-ready version of ChatGPT running on dedicated servers. They promise “enhanced security” and insist no patient data will feed back into model training. Yet the announcement raises questions: What security flaws do they claim to fix? If they won’t train on your data, can they still fine-tune behind closed doors? The article compares this to 23andMe’s quietly updated terms that let it sell DNA data after customers clicked “agree” without reading. The author points out a pattern of healthcare stunts from OpenAI. AI Consult, billed as a clinical copilot, turned out to be a human-curated ChatGPT. HealthBench was a 5,000-conversation dataset, not a new model. GPT-5’s health features were promoted using a cancer survivor’s story, then retracted amid lawsuits. Each rollout looked flashy but offered no genuine innovation—just wrappers around existing ChatGPT. Behind the scenes, OpenAI may be testing the economics of digital health. Margins in healthcare are slim for Big Tech today, so they hype these features to gauge demand. The real prize is undercutting specialized players—companies that build HIPAA-compliant systems from the ground up. If ChatGPT Health gains even a toehold, OpenAI could expand or pivot quickly, using the same marketing playbook. Bottom line: ChatGPT Health doesn’t solve core issues around data privacy or clinical reliability. It’s a way for OpenAI to signal its entry into healthcare and see how regulators, providers, and patients react. The article warns that unless you dig into the fine print, you may end up trusting a service built more on PR than on robust safeguards.