readonrisk

a16z (Andreessen Horowitz) · Oct 20, 2026 · 55 min

OpenAI’s Head of Health AI on Whether You Should Trust ChatGPT

Karan Singhal of OpenAI tells a16z's Jay Rugani how OpenAI is applying frontier AI to health: it serves hundreds of millions of weekly health users, adds specialised training to general models, and builds trust through physician networks and benchmarks. He expects falling costs of health intelligence, proactive personal health agents, and compression of administrative spend, while saying AI will not replace care delivery and that macro cost savings have not yet shown up.

Watch on YouTube ↗More from a16z (Andreessen Horowitz)

Where they stand

AI

Very bullish

AI capability and adoption in health will keep compounding, with falling costs, proactive agents and new treatments, though real macro cost savings need deployment and integration.

High conviction · years

Key takeaways

  1. More than 300 million people use ChatGPT weekly for health and wellness questions, mostly free users and mostly outside the US; OpenAI treats this as a core responsibility. 2:11
  2. OpenAI argues general frontier models with specialised health training beat small specialised models, which fall behind the frontier quickly; bet on scale. 7:07
  3. Trust is built through a global network of hundreds of physicians, evaluations like HealthBench and a mental health benchmark, and prioritising safety over user reassurance. 17:24
  4. GPT-6 Luna reportedly beats the model that was the best a month earlier on HealthBench Professional while being 34 times cheaper; health intelligence cost is expected to plummet. 44:12
  5. Healthcare spend is expected to compress over coming years, mainly through administrative and workflow integration, but this has not happened yet and abundant intelligence alone won't do it. 45:49
  6. Provider-payer AI arms race is seen as a short-term distraction that can be avoided with shared ground-truth visit data and AI adjudication. 46:54
  7. Builders are told to bet on models getting better; AI will not provide all medical care, and OpenAI will act as a quarterback while partners deliver care. 48:32
  8. Biggest impact in 5 to 10 years expected from AI-guided drug target generation and AI-guided prioritisation of clinical trials; robotics in care is not a next-few-years thing. 53:59

Calls and positions

  • The cost of a given level of health intelligence will plummet, leading to commoditisation of health intelligence.
    “the cost for that level of intelligence also dramatically reduces like it has been”
    Very bullishPrediction: dramatic cost reduction at a given intelligence levelKaran Singhal, Oct 20, 2026Watch at 45:17#OpenAI
  • Healthcare spend will see overall compression in coming years, driven by administrative and back-office workflow automation, though it has not happened yet.
    “we are really optimistic about kind of an overall compression of spend over the over the coming years”
    BullishPrediction: overall compression of healthcare spend (coming years)Karan Singhal, Oct 20, 2026Watch at 45:49
  • Clinicians and patients will increasingly feel safer with AI in the care journey, though this is not fully the case yet.
    “both parties might actually feel safer with AI in their care journey”
    BullishPrediction: both parties feel safer with AI in careKaran Singhal, Oct 20, 2026Watch at 34:56#OpenAI
  • The biggest health impact in 5 to 10 years will come from new treatments via AI-guided target generation and trial prioritisation.
    Very bullishPrediction: new and more personalised medicines (5 to 10 years)Karan Singhal, Oct 20, 2026Watch at 53:59#OpenAI
  • Robotics in care is not expected to matter in the next few years.
    “My expectation is that it's not like a next few years kind of thing.”
    BearishPrediction: robotics in care not near-term (next few years)Karan Singhal, Oct 20, 2026Watch at 51:14

Views expressed

  • A general frontier model with specialised health training is better than a separate small specialised health model, which would fall behind the frontier quickly.
    BullishKaran Singhal, Oct 20, 2026Watch at 7:07#OpenAI
  • GPT-6 Luna outperformed the prior best model on HealthBench Professional despite being 34 times cheaper.
    “it performs better despite being 34 times cheaper”
    Very bullishKaran Singhal, Oct 20, 2026Watch at 44:44#OpenAI
  • The provider-payer AI arms race is a short-term distraction that can be avoided with a shared ground truth and AI adjudication.
    “I don't think you need to have all this process where where you have like an AI arms race between payers and providers”
    BullishKaran Singhal, Oct 20, 2026Watch at 47:26
  • Entrepreneurs in healthcare should bet that models will keep getting more capable.
    “bet on the models getting better. That's probably the most important thing”
    Very bullishKaran Singhal, Oct 20, 2026Watch at 49:04#OpenAI
  • AI will not provide all medical care; OpenAI will help navigate care while partners deliver it.
    “I don't think AI can can or will provide all of that”
    NeutralKaran Singhal, Oct 20, 2026Watch at 31:10#OpenAI

Topics covered

More from a16z (Andreessen Horowitz)

Summary generated by AI from the video's captionsand checked against it; quotes are verbatim. Processed Oct 25, 2026.

Read on Risk summarises what other people said in public videos and podcasts. Summaries are generated by AI and checked by code against the transcript, but can still be wrong; follow the source links before relying on anything. This is not investment advice and nothing here is a recommendation to buy or sell. Disclaimer ·How this works