Anthropic is now a drug company. Read that sentence again.

Last week I wrote that AI is a tool, and the agency lies with us humans. I also said, quietly, that the money going into AI-for-health is not neutral money — that the labs pouring billions into "science" are not doing it out of curiosity, and that we should watch what they actually build, not what they announce.

We didn't have to wait long.

On June 30, Anthropic — the company that makes Claude, valued somewhere between "obscene" and "you can't be serious" — told STAT News it will begin developing drugs of its own. Not partnering with pharma. Not selling a model to Pfizer. Developing drugs. As a business line. Because, in their own words, "firsthand experience with Claude Science will yield benefits."

Read that sentence again. An AI lab is going to become a drug company so that its AI gets better at being an AI. The patient is the training data. The molecule is the demo.

This is the ulterior motive, uncostumed. It's not even trying to hide.

Here is what actually happens next, if you follow the incentive structure and not the press release.

An AI lab that also owns the drug pipeline gets to choose which diseases are worth solving. That choice will not be made by public health need. It will be made by whichever therapeutic area produces the most legible, patentable, high-margin molecule that also happens to make the model look brilliant in a demo. Rare oncology. GLP-1 adjacents. Anything with a ten-year exclusivity window and a billion-dollar peak. The unglamorous stuff — nutrition, sleep, movement, air, the actual determinants of health for the actual billions of humans alive — stays exactly as unfunded as it is today. Possibly more so, because now the smartest capital in the room has decided health means pharmacology.

An AI lab that also owns the drug pipeline gets to define what "evidence" means. Today the FDA is already, visibly, wrestling with what an AI-generated dossier even is. When the company running the model is also the company running the trial and also the company writing the guidance-response letter, the referee is playing three positions at once. That is not a conspiracy theory. That is a Tuesday.

An AI lab that also owns the drug pipeline gets to lock in the next twenty years of what medicine even is. Every dollar Anthropic spends on molecule discovery is a dollar not spent on the far cheaper, far more effective, far less profitable interventions — the ones that don't need a lab, a patent, or a prescription. This is the same trick the food industry pulled in the eighties: convince a generation that health is a product you buy, not a life you build, and then sell them the product. AI-pharma is going to do it again, faster, and with better graphs.

And the final one, the one I keep coming back to: an AI lab that also owns the drug pipeline is now betting its valuation on you being sick. Not on you being well. Not on capacity. Not on prevention. On chronic, medicated, monthly-refill sick. The business model requires it. The investors demand it. Claude Science will get very, very good at finding new molecules for problems that could have been solved with a walk, a plate of vegetables, and eight hours of sleep — because nobody has figured out how to charge $12,000 a month for that.

I want to be careful here. I am not anti-drug. Some of the people I love are alive because pharma did its job. Molecules matter. Trials matter. Rigour matters. If Anthropic's chemists cure a real disease, I will be the first to clap.

But we are not being asked to evaluate a chemistry breakthrough. We are being asked to accept, with a straight face, that the same company selling us an AI assistant should also be the company deciding which molecules enter our bloodstream. That's not innovation. That's vertical integration of the human body.

The counter-move is boring and it is the only one that works. Build capacity. Build the systems — food, sleep, breath, movement, meaning — that make the drug unnecessary in the first place. Refuse the framing that health is a molecule you buy from a lab that also owns the model that told you you needed it.

I said last week the agency lies with us. This is what agency looks like in practice. It looks like noticing when the tool starts selling you the disease.