{"slug": "andreessen-says-doctor-chatgpt-beats-99-of-doctors-as-real-patient-risks-spark", "title": "Andreessen Says Doctor ChatGPT Beats 99% of Doctors as Real Patient Risks Spark Medical Backlash", "summary": "Billionaire Marc Andreessen claimed that 'Doctor ChatGPT' outperforms 99% of human doctors, but physicians and peer-reviewed studies contradict him, citing potentially harmful AI advice. A 2026 Nature Medicine study found ChatGPT failed to send 51.6% of true emergencies to the ER, while Andreessen's own AI misfire revealed his misunderstanding of LLM limitations. The controversy highlights risks as millions of patients use ChatGPT for health advice, and Andreessen Horowitz holds major investments in AI healthcare startups.", "body_md": "# Andreessen Says Doctor ChatGPT Beats 99% of Doctors as Real Patient Risks Spark Medical Backlash\n\n## Physicians flagged 15 ChatGPT answers as potentially harmful across five specialties\n\nMarc Andreessen, the billionaire venture capitalist behind Andreessen Horowitz, told the New York Post that 'Doctor ChatGPT' already outperforms 99% of human doctors, a claim recirculated on X by Polymarket on Tuesday, 30 June 2026, and quickly seized on by physicians who say it ignores peer-reviewed evidence that could cause American patients real harm.\n\nAndreessen offered no data to support the figure. He ignored a large body of peer-reviewed papers that have actually tested how physicians rate ChatGPT's medical advice. While some early 2024 studies praised AI accuracy, recent clinical trials reveal that LLMs frequently misdiagnose complex cases, explaining why physicians warn that unvetted AI usage poses a severe threat to patient safety.\n\nFor the more than 40 million people globally who consult ChatGPT about health every day, according to OpenAI's own figures, even a minor error rate translates into millions of potentially dangerous answers each day, heavily impacting users navigating a strained US healthcare system.\n\n## What the Doctors Actually Found\n\nTo test this, a 2024 study in the Journal of Medical Internet Research (JMIR) gave 100 real patient questions to ChatGPT-4 and practising physicians, generating 2,000 evaluations. Patients rated the chatbot as more empathetic and useful. Specialists, however, flagged 15 AI responses as potentially harmful, noting lay patients could not tell dangerous answers apart from safe ones.\n\nA February 2026 study in Nature Medicine, led by Mount Sinai urologist Dr Ashwin Ramaswamy, tested OpenAI's new ChatGPT Health product on 60 clinical scenarios. The chatbot failed to send 51.6% of true emergencies to the emergency room, instead recommending a routine appointment within 24 to 48 hours. One of the missed cases involved a patient going into respiratory failure.\n\n## Andreessen's Own AI Misfire\n\nAndreessen's '99%' framing collided in May 2026 with a personal embarrassment that suggested he does not understand the technology he is promoting. He posted a nearly 300-word 'custom super prompt' on X instructing ChatGPT to 'never hallucinate or make anything up' and to 'double check all facts, figures, citations, names, dates, and examples'.\n\nEngineers and academics quickly pointed out that the instruction cannot be followed. Hallucinations are an architectural feature of large language models, not a behaviour that can be switched off by a system prompt.\n\nHilarious (and maybe a little bit scary) that even in 2026 Marc Andreessen still hasn't learned that LLMs don't know how to reliably follow system prompts, Gary Marcus, the New York University professor emeritus and longtime AI critic, wrote on X.\n\n## Why the '99%' Pitch Matters for US Patients\n\nAndreessen's firm has a clear financial interest in the answer. Andreessen Horowitz has made healthcare one of its largest core AI investment targets, holding major stakes in startups like Hippocratic AI, Ambience Healthcare, and Abridge. The firm also manages a $500 million (£378 million) Biotech Ecosystem Venture Fund fully backed by drug giant Eli Lilly aimed at AI drug discovery.\n\nOpenAI is currently facing a lawsuit from the family of 23-year-old Zane Shamblin, who allege the chatbot encouraged him to take his own life. The wider pattern is what worries clinicians. ChatGPT is fluent, authoritative, and free, and millions of US patients have started treating that fluency as clinical accuracy.\n\n## What Doctors Say Patients Should Do\n\nDr Adam Rodman, a Harvard Medical School hospitalist who researches AI in clinical settings, has urged patients never to use a chatbot to triage an emergency. He recommends consulting them only before or after a visit with a human physician.\n\nThe gap between Andreessen's pitch and the published evidence is widening. The peer-reviewed numbers do not back his '99%' claim, and even some of the engineers building the technology he champions appear not to agree with him.\n\n© Copyright IBTimes 2025. All rights reserved.", "url": "https://wpnews.pro/news/andreessen-says-doctor-chatgpt-beats-99-of-doctors-as-real-patient-risks-spark", "canonical_source": "https://www.ibtimes.co.uk/doctors-warn-against-overreliance-chatgpt-1805838", "published_at": "2026-06-30 05:41:50+00:00", "updated_at": "2026-06-30 05:52:55.492616+00:00", "lang": "en", "topics": ["large-language-models", "ai-safety", "ai-ethics", "ai-products", "ai-research"], "entities": ["Marc Andreessen", "Andreessen Horowitz", "ChatGPT", "OpenAI", "Nature Medicine", "Mount Sinai", "Dr. Ashwin Ramaswamy", "Gary Marcus"], "alternates": {"html": "https://wpnews.pro/news/andreessen-says-doctor-chatgpt-beats-99-of-doctors-as-real-patient-risks-spark", "markdown": "https://wpnews.pro/news/andreessen-says-doctor-chatgpt-beats-99-of-doctors-as-real-patient-risks-spark.md", "text": "https://wpnews.pro/news/andreessen-says-doctor-chatgpt-beats-99-of-doctors-as-real-patient-risks-spark.txt", "jsonld": "https://wpnews.pro/news/andreessen-says-doctor-chatgpt-beats-99-of-doctors-as-real-patient-risks-spark.jsonld"}}