Welcome to AI Health Uncut, a brutally honest newsletter on AI, innovation, and the state of the healthcare market. If you’d like to sign up to receive issues over email, you can do so here.
Important Disclosure. This publication is written and distributed by an independent journalist. It is protected by the First Amendment to the U.S. Constitution and related principles of free expression. Those protections do not relieve me of the obligation to report accurately, and I take that obligation seriously. I have no financial interest, long or short, in Nvidia, OpenAI, Anthropic, Alphabet, xAI, Meta, Insilico Medicine, Recursion, Isomorphic Labs, Novartis, or any other company mentioned here. This article is for informational and opinion purposes only and should not be construed as financial or investment advice.
🚨 But first, a quick personal note…
Last week I spoke at the inaugural Origin Point conference in La Jolla, and I’ll be as candid about it as I am about everything else: it was superb. About 85 people, lots of investors, lots of founders, and small enough that you could have real conversations instead of collecting badges. We covered everything from longevity, GLP-1s, peptides and biomarkers to consumer healthcare AI, wearables, AI agents, women’s health, M&A, and, fittingly for this issue, AI’s potential role in curing cancer. Thank you to the organizers, Noah Voreades, Callie Baker and Jacqueline Jacques, and to our panel moderator Julia Croxen and my fellow panelists Arpan Parikh, MD MBA FAPA and Anson Whitmer for a lively discussion on digital health, consumer health and AI products. Also, black caviar for breakfast. 😉 I posted more on LinkedIn. If you get the chance to attend next year, definitely do it.
By the way, if you cannot afford this article, perhaps you’re a student or currently between jobs, please reach out. That’s precisely why I created the AI Health Uncut Founding Member Club. Thanks to generous donations from these wonderful individuals, I’m able to provide access to anyone who needs it.
On September 6, Jensen Huang posted that OpenAI’s new GPT-6 Astra means “AGI has arrived.” Three days later, Tanishq Mathew Abraham launched CancerBench, a leaderboard with one metric: how many types of cancer has your frontier model cured? GPT-6 Astra scored 0. So did Claude Fable 5.1, Gemini 3.8 Flash, Grok 4.6, and Muse Spark 1.3. “All tied for first. And last.”
Is it AGI if it scored exactly zero on CancerBench? 🤔
On September 7 I published There Is No Money in AI Drug Discovery. The most common follow-up question was about cancer. This is the answer.
TL;DR
CancerBench: 5 frontier models, 0 cancers cured, days after Nvidia’s CEO declared AGI. The promises behind the joke were serious: Amodei’s “~5-10 years,” Hassabis’s “next decade,” Altman’s “[choice] to cure cancer.”
AI-discovered molecules clear Phase 1 at 87.5% (21 of 24), against 52.0% historically. In Phase 2 they clear at 40% (4 of 10), no better than anyone else. None has finished a Phase 3. For cancer, the AI-specific Phase 2 record is 0 trials.
Oncology, the largest field in drug development, has among the worst odds: a 5.3% chance of approval from Phase 1, against 7.9% for all drugs, and a 24.6% Phase 2 pass rate, its worst phase.
Mayo Clinic’s AI-designed pancreatic cancer drug, GIPCi, is real, preclinical, and not yet in Phase 1. From there, the average oncology clock to approval is 10.3 years.
Human trials cannot be simulated, and the FDA has cleared exactly 0 generative AI products and has no pathway for one. My estimate: 10 to 20 years, not 5, and probably not with transformer AI.
Here is what’s inside:
1. Five Frontier Models. Zero Cancers Cured.
2. AI Fixed Phase 1. Phase 2 and 3 Have Not Moved.
3. Mayo Clinic’s AI Cancer Drug Is Real. It Is Also Preclinical.
4. Human Trials Cannot Be Simulated, and the FDA Is Not Ready to Try
5. The Potemkin Dashboard and the Naked Emperors
6. Expect 10 to 20 Years, Not 5
