Reid Hoffman told doctors at WIRED Health in London on April 16, 2026 that any clinician not consulting a frontier AI model for a second opinion is 'bordering on committing malpractice.' The LinkedIn cofounder, who sits on the boards of PayPal and OpenAI and now runs the cancer drug discovery startup Manas AI, framed the argument as a patient safety issue rather than a productivity pitch. He said he personally runs his own medical questions through frontier models and expects his concierge doctors to do the same.
His reasoning rests on scale. Frontier models from companies including OpenAI and Anthropic, Hoffman argued, have ingested trillion-plus words of information across medicine, biology, and case literature, even when they were not specifically trained for clinical use. 'As a second opinion, it is bringing superpowers that no human being has,' he told the audience.
Hoffman was careful to draw a line between using a model as a check and outsourcing the diagnosis to it. 'You could very well go, no, I think you're wrong, I think it's this,' he said. 'But if you're not using this as a second opinion, you're making a mistake, both as a doctor and as a patient.'
Key facts
- 01Reid Hoffman told WIRED Health in London on April 16, 2026 that doctors not using frontier models as a second opinion are 'bordering on committing malpractice.'
- 02Hoffman's drug discovery startup Manas AI, cofounded with oncologist Siddhartha Mukherjee, aims to compress drug discovery from a decade to a few years.
- 03He wants the FDA to run biological-model tests to fast-track promising drugs, but said meaningful adoption is not coming soon.
- 04Hoffman predicts that within 10 years every major disease will have target molecules capable of making a serious difference.
- 05His three-decade Silicon Valley resume includes cofounding LinkedIn and board seats at PayPal and OpenAI.
The framing will land awkwardly with much of the medical profession. A major study earlier in 2026 concluded that large language models pose real risks to members of the public seeking medical advice, citing inaccurate and inconsistent answers across repeated prompts. Hoffman's response is that the alternative for most patients is no specialist input at all, not a perfect one.
“Frontier models have ingested trillion-plus words of information, Hoffman argued, and within 10 years he expects every major disease to have target molecules that make a serious difference.”— Jaeden Schafer
He pointed to the United Kingdom's National Health Service, which is straining under long waiting lists and a chronic shortage of family doctors, as the obvious test case. 'We just don't have enough doctors, most people don't have access, and when you think about, how should the NHS be redesigned, everyone should be interacting with this medical assistant,' Hoffman said. He pitched a free, smartphone-based assistant that could triage cases before they reach a human clinician.
Hoffman has a clear commercial interest in AI taking a bigger role in medicine. Manas AI, which he cofounded with the oncologist Siddhartha Mukherjee, is building an engine to identify cancer drug candidates and explicitly aims to shift drug discovery from a decade-long process to one that takes a few years. Mukherjee personally reviews the system's proposals, Hoffman said, separating real leads from what he called the 'bonkers stupid' ones.
He also wants regulators to use AI on the other side of the desk. 'As a Silicon Valley person, I would love to get to a point where the FDA was also running tests with biological models, going, oh, we should fast-track this one, because the likelihood of negative consequences is lower,' Hoffman said. Asked whether that was imminent: 'Unfortunately, no.'
The longer-term claim is bolder. 'I think in 10 years, every major disease will have target molecules that could at least make a serious difference,' Hoffman said, including chronic and rare conditions that have historically been uneconomical for pharmaceutical companies to chase. Manas is starting with cancer, but Hoffman frames the discovery engine itself as the asset.
The skeptics' case is straightforward and unresolved. Frontier models still hallucinate, change their answers between sessions, and have no licensed pathway as clinical decision support in most jurisdictions. A doctor who follows a chatbot into a misdiagnosis has far less legal cover than one who follows another physician, and Hoffman did not address where liability lands when the second opinion comes from GPT or Claude.
Hoffman's pitch is really an argument about distribution. The bottleneck in global health is not the quality of the best specialist; it is access to any specialist at all, and frontier models from OpenAI and Anthropic are already in roughly a billion pockets. Whether the medical establishment treats that as a tool or a threat will shape the next decade of how AI shows up at the bedside, and Hoffman, with one foot in OpenAI's boardroom and another in a drug discovery startup, is not a neutral narrator. He is, however, an unusually well-positioned one.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
