Singhal says asking five physicians one question can yield eight answers, so OpenAI built a network of hundreds of experts across specialties, languages, and countries. That network shapes evaluations and training data instead of letting one or two clinicians adjudicate hard cases. It is a deliberate aggregation of expert judgment.
Singhal says OpenAI wants to help patients navigate scheduling, logistics, and care questions, but it will rely on startup and provider partners for the last mile. He describes ChatGPT as a quarterback for health rather than a replacement for care delivery. The goal is enabling the ecosystem.
Singhal says healthcare fragmentation and disconnected workflows block AI adoption, and legacy EHRs often run on cloud virtual machines. Computer-use agents could bridge those systems, though he warns not every healthcare app will have a clean connector soon. That makes agents especially promising for interoperability.
Singhal says EHR histories, claims, and other stored healthcare data hold insights that are not fully mined. He sees agents helping generate new knowledge from that data and accelerating validation, which may become the key bottleneck in medical research. Scientists are a third core constituency.
Karan Singhal says OpenAI adds specialized healthcare data to every training phase for ChatGPT, rather than forking a smaller medical-only model. A separate model would fall behind the frontier too quickly, so OpenAI hitches health expertise to scale and general reasoning improvements. That is the best-of-both-worlds approach he describes.
Singhal cites studies finding clinical guidelines can take five to 17 years to penetrate everyday practice. OpenAI's clinician products try to compress that lag by bringing latest biomedical literature into the point of care, including during a 15-minute visit. That lag is a massive bottleneck.
Singhal says GPT-6 Luna beat the prior best model on HealthBench Professional while being 34 times cheaper. OpenAI expects the cost of a fixed level of trustworthy health intelligence to keep dropping, making abundant deployment more feasible. That is central to its access mission and pricing bets.
Singhal says users may want comforting answers, but evals track cases where the model must say the situation is unsafe and the user should seek care. OpenAI treats this as a safety prioritization across training and evaluation, even when users resist the message. That remains central as health usage scales.
Singhal says medical care bundles diagnosis, physical exams, empathy, and legal liability that AI cannot fully provide. Even with pre-visit research, the human visit remains necessary for history-taking, physical interaction, comfort, and a second check. That limits AI's role to support rather than replace care.
Singhal says OpenAI focused on security, HIPAA compliance, and privacy so healthcare enterprises could adopt ChatGPT. Epic integration and a sidebar inside the EHR reduce the activation energy for clinicians and keep workflows from fragmenting across tabs. This matters as much as model quality for adoption.
Singhal says Mental Health Bench goes past baseline safeguards to measure whether models provide positive mental health experiences. Licensed experts worldwide helped define tricky scenarios because mental health has even more disagreement than physical health. The benchmark is meant as a shared ruler for the industry.
Singhal invokes AlphaGo's move 37 to describe future health insights that initially surprise experts but later look brilliant. He expects models to integrate years of EHR, chat, and claims data to flag risks humans cannot hold in their heads. Panda Health's antibiotic overuse flag is an early example.