OpenAI's ChatGPT Health tools spark safety debate

Catch up quick: OpenAI is adding a new health tab within ChatGPT that includes the ability to upload electronic medical records and connect with Apple Health, MyFitnessPal and other fitness apps.

• The new features formalize how many people already use the chatbot: uploading test results, asking questions about symptoms and trying to navigate the complex health care ecosystem.
• OpenAI says it will keep the new health information separate from other types of chats and not train its models on this data.

Yes, but: Health information shared with ChatGPT doesn't have the same protections as medical data shared with a health provider, and even those protections vary by country.

• "The U.S. doesn't have a general-purpose privacy law, and HIPAA only protects data held by certain people like healthcare providers and insurance companies," Andrew Crawford, senior counsel for privacy and data at the Center for Democracy and Technology, told Axios in a statement.
• "And since it's up to each company to set the rules for how health data is collected, used, shared, and stored, inadequate data protections and policies can put sensitive health information in real danger," he says.
• OpenAI is starting with a small group of early testers, notably not those in the European Economic Area, Switzerland and the United Kingdom where local regulations require additional compliance measures.

The big picture: Many AI enthusiasts on social media welcomed the tools, pointing to how ChatGPT already helps them.

• Yana Welinder, head of AI for Amplitude has been using ChatGPT "constantly" for health queries in recent months, both for herself and family members. "The only downside was that all of this lived alongside my very heavy other usage of ChatGPT," she wrote on X. "Projects helped a bit, but I really wanted a dedicated space …So excited about this."

Chatbots aren't a replacement for doctors, OpenAI says, at the same time highlighting what the chatbot can provide.

• "It's great at synthesizing large amounts of information," Fidji Simo, CEO of applications at OpenAI, said Wednesday on a call with reporters. "It has infinite time to research and explain things. It can put every question in the context of your entire medical history."

The other side: AI skeptics question giving medical information to a chatbot that has shown a propensity to reinforce delusions and even encourage suicide.

• "What could go wrong when an LLM trained to confirm, support, and encourage user bias meets a hypochondriac with a headache?" Aidan Moher wrote on BlueSky.
• Anil Dash, advocate for more humane technology, agreed on BlueSky that it "isn't a good idea," but also wrote that "it's vastly more understandable than most medical jargon, far more accessible than 99% of people's healthcare that they can afford, and very often pretty accurate in broad strokes, especially compared to WebMD or Reddit."

Between the lines: Like other information shared with ChatGPT, health information could potentially be made available to litigants or government agencies via a subpoena or other court order.

• That seems particularly noteworthy at a time when access to reproductive health care and gender-affirming care are under threat at both the state and federal levels.
• User data could get swept up in other ways, too. As part of their copyright battle against OpenAI, news organizations have obtained access to millions of ChatGPT logs, including from temporary chats that were meant to be deleted after 30 days.
• Sam Altman has called for some sort of legal privilege to protect sensitive health and legal information.

What to watch: OpenAI said it has more health features on its road map and will talk soon about additional work with various health care systems.