·6:38

Anthropic's Record IPO and Google's $5B AI Drug Bet — Nov 20, 2026

Show notes

Five billion dollars for AI drugs, and not one has reached a patient yet.

Run time: 6:38

In today's episode:

  1. Isomorphic raises $5B; still no drug in human trials
  2. Vanderbilt AURA trial cuts autoimmune diagnosis wait 138 to 43 days
  3. AlphaGenome 2 doubles causal-variant hits, per company preprint
  4. FOLLOW-UP: AHA's WATCH-LV: smartwatch brands disagree on 9%
  5. OpenAI sends computer-use agents into ChatGPT for Healthcare
  6. Apple and OpenEvidence open 200,000-person health AI study
  7. FDA swaps "artificial intelligence" for "super intelligence"
  8. GE clears on-device feeding-tube AI for portable X-ray
  9. FOLLOW-UP: NEJM AI chest X-ray fracture model, densitometry-referred caveat
  10. Anthropic prices record $81B IPO, trades as ANT
  11. Google ships Gemini 4 Neon at $3/$15 per million tokens
  12. OpenAI GPT-6.2 lands to mixed early-tester reviews

TL;DR:

  • Isomorphic Labs raised $5B (Nov 17) on top of a $2.1B round six months ago. No candidate has entered human trials, and every result so far is preclinical and self-reported.
  • AURA (NEJM AI, Nov 6) is the week's real evidence: 2,385 ANA-positive patients randomized at Vanderbilt. Median time to diagnosis fell from 138 to 43 days (HR 1.9).
  • Anthropic went public on Nov 19 at $83 a share, about $81.3B of stock. Its company-published biology claims now sit under public-market incentives, so outside replication matters more.

Sources cited:

Subscribe: YouTube

medAI Times is for educational and informational purposes only. The content does not constitute medical advice, diagnosis, treatment recommendation, or professional clinical guidance. Consult qualified healthcare professionals and refer to official sources before making clinical, research, regulatory, or business decisions.

Transcript

Auto-generated from the episode audio. Click any timestamp to jump the player there.

$5 billion for AI drugs, and not one has reached a patient yet. Welcome to MedAI Times Podcast, your daily update on medical AI. Don't forget to like and subscribe. Isomorphic Labs raises $5 billion.

Anthropic goes public in the largest U.S. listing on record. A Vanderbilt trial gets lupus-type diagnoses made three months sooner. Google ships Alpha Genome 2. And the FDA renames artificial intelligence.

On Wednesday, a chest X-ray predicted who would break a bone. This week, a Vanderbilt trial showed what happens when a prediction actually changes who gets seen. Top Story. On November 17, Isomorphic Labs, the Alphabet drug discovery company that grew out of DeepMind's AlphaFold work, closed a $5 billion Series C.

Green Oaks and Sequoia led. Andreessen Horowitz, Alphabet, Thrive, and the UK sovereign AI fund joined. That comes about six months after a $2.1 billion round, which puts total outside money above $7.7 billion.

Demis Hassabis says the money will push its first therapeutic programs toward clinical trials in oncology and immunology, alongside partnerships with Eli Lilly, Novartis, and Johnson & Johnson. Now the plain part.

PharmaForum's report states that Isomorphic has yet to move a single drug candidate into human trials. There is no trial date. Every result so far is preclinical and self-reported. We've covered the bigger pattern before.

AI-designed molecules tend to clear phase 1 safety at high rates, then fail on efficacy about as often as everyone else's. So here's our verdict. For clinicians, this is noise this year. $5 billion buys compute and chemists.

It doesn't buy a patient result, and that's the only scoreboard at the bedside. The story every newsroom led with. Anthropic priced its initial public offering at $83 a share on November 18th and started trading on the NASDAQ under the ticker ANT the next day.

980 million shares were sold, about $81 billion of stock. Renaissance Capital calls it the largest US IPO on record. Simaphore reports shares touched $111 intraday.

Of those shares, $956 million came from Anthropic and the rest from existing holders. Why should a medical channel care? Over the past six weeks, Anthropic has put out a string of biology claims.

Claude designed Scorpion toxin binders, a drug program run with 11 outside labs, work on vault particles. All of it was company-published, and most of it is not peer-reviewed. Public shareholders now have a financial reason to cheer those claims.

So independent replication matters more than it did a month ago. Watch who outside the company reproduces them. The best evidence of the fortnight came from Vanderbilt. The URA trial in NEJMAI randomized 2,385 patients with a positive ANA test.

In half of them, a risk model built into the health record flagged high-risk patients for a fast rheumatology visit. In the other half, the score ran silently. Among patients who were diagnosed, the median wait from positive test to diagnosis fell from 138 days to 43.

One caveat, the share diagnosed within six months, 23 versus 16 percent, missed significance. It's one health system. It's still a real randomized trial in real care. On November 11th, Google DeepMind released Alpha Genome 2.

It reads a million letters of DNA and predicts gene activity for a specific tissue or cell type trained on more than 100 million single cell profiles. In 28 immune cell types, it found 30 percent of known causal variants, compared with 16 percent for the previous version.

Those figures come from the company's own preprint. It's labeled research use only, with no clinical validation. A follow-up from the American Heart Association meeting, we covered two AI trials there as shorts. The third worth knowing is WATCH-LV, a Korean study of 2,211 adults.

AI read 30-second smartwatch ECGs to flag a weak heart pump. Apple and Samsung watches disagreed on 9 percent of people, requiring a positive reading on two of three days halved false alarms, but it missed two more real cases.

It's an abstract, not peer-reviewed, with no outcome data. OpenAI pushed ChatGPT for healthcare into hospital-backed offices on November 16th. The update adds computer-use agents that click through existing software, plus more than 20 plugins for coding, insurance eligibility, and pricing.

CVS, Cigna, Abridge, and Ambience are the first named users, working on referrals, prior authorization, and discharge planning. The safety figures OpenAI cites are physician ratings of answers.

None of them measure patient outcomes. At HLT8 in Las Vegas, Apple launched a study with open evidence on how people use AI for health questions. It will enroll up to 200,000 U.S. adults for at least a year, and Apple says the data won't train models.

There are no results yet. It's a question being asked at scale, not an answer. And the regulatory note of the week. Under a September executive order, the FDA now says superintelligence instead of artificial intelligence in its own documents.

Existing rules don't change, and device makers don't have to follow. A Duke policy researcher called it a semantic change. The software, it turns out, did not get smarter overnight. One clearance to note? GE Healthcare received FDA clearance for Critical Care Suite 3.0, which spots feeding tubes on portable chest X-rays in adults and children without a cloud connection.

The trade coverage reports 99.2% sensitivity. We couldn't reach GE's own release to confirm that number. Spotlight. Time-to-event analysis. Aura didn't just count who got diagnosed.

It tracked how long each patient waited and compared the curves, which gave a hazard ratio of 1.9. That means patients in the alert arm were diagnosed at roughly twice the rate at any given moment. Every study and filing from this week is linked in the description, with its evidence tier next to it.

Our question for you this week. Would you enroll one of your patients in a trial of a drug designed mostly by an AI model? Thanks for listening. Find us on YouTube and your favorite podcast app. See you tomorrow.