The Pivot
Earlier this week Anthropic dropped a new tool called Claude Science at “The Briefing: AI for Science.” It’s an “AI workbench,” a place to drag in messy datasets and fragmented tools and spit out clean visuals. Standard tech bro pitch stuff. Except there was a kicker.
Eric Kauderer-Abrams their head of life sciences said something odd. He said Anthropic intends to develop drugs itself.
Not just the software. Not just the consulting. Drugs.
Specifically for neglected diseases.
“AI has the potential to dramatically accelerate the pace of scientific discovery.”
They already dominate coding. They have powerful models. They’ve got pharma clients lining up. Now they’re entering the ring as a competitor too? That’s bold. Or stupid. Hard to say which.
The Vague Promise
Here’s the catch. Nobody really knows how this works.
Anthropic hasn’t spilled any details. Which diseases? Who are the partners for animal testing? How about manufacturing? Kauderer-Abr Abrams stayed silent on what happens if they actually find a lead. The Verge asked. Silence answered back.
It feels like a bigger fog surrounds the whole “AI drug” concept anyway. Namshik Han a Cambridge professor put it simply: the term is too broad. AI touches everything now.
From finding a new compound.
To analyzing trial data.
To manufacturing logistics.
Everyone uses it. It’s a catch-all phrase because the tech is everywhere.
Hype vs. Reality
Is AI changing the game? Sure. It’s fast.
Giants like AstraZeneca and GSK are using it to brainstorm. To suggest new molecules that might hit cell receptors. Matthew Todd from UCL says it’s great for speed. For “road testing” ideas before spending millions. Anthropic’s generative models could scan vast biological oceans for needles we missed.
But finding a needle is different from stitching a coat.
We are far from an AI-designed drug in your pillbox. Todd warns we’re still years off. Decades even. Because you can’t automate the hard parts.
Experiments are still required. Real ones. Not simulated ones.
“They haven’t yet come close to making experiments necessary.” — Frank von Delft
You need to test for toxicity. You need to check if it degrades in a stomach. You need humans to take it and see if they don’t die. That takes skilled labor. Money. Time. Lots of it. If Anthropic wants a drug on the market they’re going to burn through cash on wet lab work.
The Human Element
So they’re building labs.
In the last year Anthropic has hired biologists. They’ve posted live roles. Namshik Han says he’s seen colleagues approached directly. Some came over. From Big Pharma. From elite academics.
It’s a quiet raid.
But here is the rub. Even if they nail the science. Even if the AI points exactly to the right molecule.
Clinical trials take time. A decade is normal. No AI drug has crossed the finish line to the FDA yet. Some candidates have started the marathon but we can’t tell if AI actually helped or just got credit.
Speed matters. But safety matters more.
AI might find the shortcut. But the body? The body still follows the old rules. Slow. Methodical. Unforgiving.
Will Anthropic survive the lag? Or will it just become another expensive experiment itself?
