A response to Dario Amodei on AI & clinical trials

Ruxandra Teslo · 28 Feb 2026

Others here said this too

About ninety per cent of drugs that enter clinical trials fail.

Their wordsThe current success rate for a drug entering clinical trials is only about ten percent, meaning 90 percent of all drugs fail.

Also said by Derek Lowe

Read the piece on writingruxandrabio.com ↗

5 korrents from this piece ↓

Ruxandra Teslo did not write this page.

Every claim below was made in this piece, quoted word for word and numbered in the order the piece makes them, so you can read it there rather than take our word for it. The sentence above each quote is our reading of the claim, not their wording. Each quote was checked against a stored copy of the page at build time; where the two differ, the quote is the fact.

Our reading
About ninety per cent of drugs that enter clinical trials fail.

Their wordsThe current success rate for a drug entering clinical trials is only about ten percent, meaning 90 percent of all drugs fail.
Our reading
Better drug candidates do not shorten clinical trials, because a trial's duration is set by operational and regulatory friction rather than by how good the drug is.
Mental modelBottlenecks

Their wordsEven if AI can help design more effective drugs, timelines will not compress until we solve the operational and regulatory bottlenecks of trials.
Our reading
A clinical trial is not only a test of a drug: early trials are experiments whose real product is the human biology data everything downstream is built on.

Their wordsTrials serve two distinct functions: validation — confirming whether a drug works and is safe — and learning, or generating biological data to refine our understanding of a disease, a compound, and the relationship between the two.
Our reading
How long a trial takes decides which diseases get drugs at all, because capital flows to the indications whose endpoints can be measured quickly.
Mental modelIncentives

Their wordsLastly, the duration of a clinical trial does not merely determine how fast an individual therapy reaches patients. It also shapes which diseases attract serious investment and which do not.
Our reading
The human data an AI model would need to predict trial outcomes can only come from trials in people, so the loop runs no faster than the trials themselves.
Mental modelBottlenecks

Their wordsBut remember that such data can only come from trials in people (mice are nice, but most animal results simply do not translate.)

Keep reading

The other pieces are the ones whose claims are worded most like this one's. That measures language, not agreement: they may argue the opposite.