The INSANE Secret Plan of AI CEOs (not what you think)

In October, Dave Blundin, a venture investor who co-hosts the Moonshots podcast, offered a scenario. Picture Moderna CEO Stephane Bancel with no AI team yet, thinking: "we'll just license from Anthropic for a few years and then we'll figure it out."

Blundin's warning: Anthropic already opened its wet labs and is working two generations of model ahead. "They're never going to release those to you. They're going to use those to develop drugs and then they're going to sell drugs."

So are the AI labs turning into medicine companies? And what happens to the medicine your family takes if the smartest AI never leaves the building? Nothing in the video, or in this piece, is investment or financial advice.

Not a chatbot company

Most people think of OpenAI and Anthropic as chatbot companies. They build the AI and sell it to anyone who pays. The drug companies are just their customers. Blundin sees it differently: the labs will keep their best models in house and sell the drugs themselves.

Two clues at Anthropic

The first clue is a wet lab, where scientists run experiments on real cells and chemicals. Reuters reported it on September 18th, and Anthropic showed its first result on September 23rd. It's in the Bay Area, and human scientists do all the hands-on work with low-risk material only. Anthropic says the lab is for basic biology, and it told Reuters the lab isn't for drug discovery specifically.

For its first result, about 950 Claude agents, copies of the AI working on their own, ran for about 21 hours. One turned up a new enzyme system that looks a lot like CRISPR, the bacterial system scientists turned into a gene editing tool. What it does is still unknown.

The second clue is bigger. In June, Eric Kauderer-Abrams, Anthropic's head of life sciences, said, "We've decided to start running some drug programs ourselves." Anthropic runs a small number of them, all preclinical, meaning before any testing in people, and all aimed at neglected diseases.

In April, Anthropic bought a small biotech startup called Coefficient Bio. That same month, Novartis CEO Vas Narasimhan joined Anthropic's board. That's a company putting together the parts a drug company needs.

Shovels and a map

Why would an AI company do this? Think of a gold rush, where the reliable money went to the people selling shovels. That's the AI labs right now. They sell the shovels, the AI by the word, to the miners, the drug companies. But what if the shovel seller figured out it also had the best map to the gold? Do you keep selling shovels or go dig? Do you sell the map or keep it?

Look at the gold. Merck's cancer drug Keytruda sold $31.6 billion in 2025 alone. The software side of medicine gets crumbs. Takeda paid $4 billion for a drug that Schrödinger's drug design software helped create, and Schrödinger got about $147 million of it.

Selling shovels is about to get harder, too. AI models are getting good enough for most everyday tasks, and then people go with whatever's cheapest and fastest. In medicine, being the smartest is worth billions, because the prize is a drug you own under a patent. The way I see it, the math pushes the shovel seller to start digging.

The best map stays home

Access to Anthropic's most powerful model, Mythos, remains limited to a small set of vetted organizations. Anthropic's public Fable models will continue to block professional biology and drug development queries because of potential dual use risks, meaning the same skill can help or harm. Yet Anthropic says, "Using Mythos 5, our internal protein design experts accelerated aspects of the drug design process by around 10 times."

To be fair, vetted drug and biotech researchers can now get Mythos through a program Anthropic opened in September. Both labs say they hold models back for safety. OpenAI sells its biology model, GPT Rosalind, only through trusted access. No lab boss has said they keep models in house to own drugs. That read comes from investors like Blundin, and I think it holds up, because the best map is worth far more kept than sold.

OpenAI has no known wet lab and no drug programs, and neither lab is listed as a sponsor on any clinical trial. But OpenAI's leaders have floated a royalty, a cut of every sale. CFO Sarah Friar put it this way on a podcast in January: "you have a breakthrough, that drug takes off and we get a licensed portion of all its sales."

Isomorphic Labs, Google DeepMind's sister company, goes further. Isomorphic runs its own drug programs, and its deals with Lilly and Novartis are worth nearly $3 billion in potential milestone payments. But it hasn't dosed a single human yet, and its first trial target slipped from the end of 2025 to the end of 2026.

Trials are the slow part

Even the AI lab that's furthest along is stuck waiting on the slowest part of medicine. But Anthropic CEO Dario Amodei was honest about the limit: "This will not in itself speed up clinical trial times." It could greatly increase the number of promising candidates, Amodei said.

Phase 1 is the first small test in people. It takes about 10 and a half years on average to get from phase 1 to approval, and only about 7.9% of drugs entering phase 1 get approved. The FDA still hasn't approved a drug designed by generative AI. AI fills the pipeline faster. People still wait on the trials.

The case against

There's a strong argument against all this. On that same Moonshots panel, Alex Wissner-Gross argued the biggest money is in recursive self-improvement, which is AI building better AI. Asked when the labs would stop releasing their next models and use them for their own applications, Wissner-Gross said, "We're already past that point." Other uses, like curing the top 5,000 diseases, are "all just marketing," a way to keep "social permission to focus on recursive self-improvement."

Wissner-Gross has a point, because better AI makes everything else better. But marketing doesn't usually come with a wet lab, its own drug programs, a biotech purchase and a drug company CEO on the board. So I think both things are true. Most of the money goes into better AI, and some of that AI gets pointed at drugs the lab owns.

Anthropic pushes back. Kauderer-Abrams said, "We're not competing with pharma and biotech companies that make their business in bringing drugs to market." On October 1st, Anthropic said it's drawing the line at phase 1 for its own programs. But Amodei also wrote in September, about drug discovery and clinical trials, "We are also interested in these later stages." My bet is that line moves as the money gets bigger.

Rivals, families, fights

Right now the big drug companies are the labs' customers. But the day the shovel seller starts digging, the miners turn into rivals. Pharma's answer is data. Pfizer CEO Albert Bourla put it this way: "Anyone can license a model. No one else has our data." As I see it, pharma has the data, the labs have the AI, and the winners need both.

For families, there are about 10,000 rare diseases, and only about 5% have an approved treatment. Anthropic says it's aiming its own programs at neglected diseases the market overlooks. But better ways to find drugs don't automatically mean cheaper drugs, and nothing so far shows AI savings bringing US prices down.

Then come four fights. First, pharma against the labs. PhRMA, the drug industry's main lobbying group, spent a record $38.2 million on federal lobbying in 2025.

Second, who gets the map. On October 6th, Blundin posted on X that if you wanted to script the AI revolution as badly as you possibly could, "You'd develop AI right up to the point where it can help somebody build a weapon, but not quite far enough to cure cancer." He argued that's where we are right now. Limiting the best biology AI to trusted groups makes sense for safety. But that same gate lets a lab keep the best map for itself. No US antitrust regulator has acted on AI labs moving into drugs yet. I think that will likely change the moment a lab owns a hit drug.

Third, the FDA. Amodei warned in June, "Without reforms, AI will simply jam or overload this system." The FDA's own guidance on AI and drug development is still only a draft.

Fourth, patents. Courts and the US patent office say an inventor has to be a human. So a lab can patent a drug its AI designed as long as a person counts as the inventor. In this gold rush, regulators decide who's allowed to dig and who keeps the gold.

The end of disease

Demis Hassabis was asked on 60 Minutes about the end of disease. Designing just one drug takes 10 years and billions of dollars on average, Hassabis said, and "We could maybe reduce that down from years to maybe months or maybe even weeks." Then: "I think one day maybe we can cure all disease with the help of AI." On timing: "Maybe within the next decade or so. I don't see why not." That's Hassabis's projection.

Here's my read. The leading AI labs are turning into medicine companies, because their best model is worth far more finding drugs they own than selling answers by the word. The story about AI killing us all gets almost all the attention, while the cure story mostly gets ignored. I think the way through this fight is to make it undeniable, with real cures reaching as many people as possible.

Go back to Blundin's scenario: a drug company CEO who plans to license from Anthropic for a few years. That plan treats the AI lab like a shovel seller. But the shovel seller now runs its own lab, and it's holding the best map it's ever had. Who gets to use that map is going to decide how fast the next cure reaches you and the people you love.