Isomorphic Labs president Max Jaderberg argues that AI could help drug researchers model human biology more directly, rather than relying so heavily on trial and error. He says rat tests predict human clinical toxicity only about 30% of the time, a gap that helps explain both the promise and the limits of AI drug discovery.
Isomorphic Labs Watch analysis
What happened
In an interview on the No Priors podcast, recounted by BigGo Finance, Jaderberg said drug development now costs roughly $1.5 billion and takes about 15 years, while annual approvals have stayed roughly flat despite a much larger scientific literature and more types of medicine.
His proposed route is to combine predictive AI models with human-relevant laboratory methods such as organoids and organs-on-a-chip. Isomorphic’s stated aim is to use models to make lab work more about validating predictions than discovering problems through repeated experiments. Jaderberg says the company measures progress against experimental accuracy, not just model benchmarks.
Key findings
- Animal tests have limits
Jaderberg estimates rat studies predict human clinical toxicity about 30% of the time; he argues that this leaves important risks to emerge in human trials. - Failures drive the long timetable
The interview describes toxicity, off-target effects and harmful effects on the intended target as recurring reasons drugs fail in clinical development. - Models are not the whole answer
Jaderberg proposes combining AI with in-vitro tests and newer human-relevant platforms, while acknowledging that data from human trials remains limited. - The company keeps its engine in-house
Jaderberg says Isomorphic uses its drug-design models internally rather than selling or licensing them.
Why it matters
The practical case for AI here is not that a model can declare a drug safe. It is that better predictions might help researchers spot problems earlier and choose which candidates deserve costly experiments. If that works, it could change where time and money go in a process where many failures arrive late.
But the 30% figure is Jaderberg’s claim in the interview, not independent validation of Isomorphic’s models. Animal tests can be imperfect without making a computer prediction ready for patients. The crucial test remains whether a model’s predictions improve experimental and clinical outcomes.
Our read
This is a sharper argument than the familiar promise that AI will simply discover medicines faster: drug development needs better ways to predict what happens in people. The company’s own claims will need evidence from real programmes, not just promising model scores. Biology, as ever, keeps the final vote.
What to watch
- Whether Isomorphic publishes evidence linking model predictions to better experimental results.
- How it combines AI predictions with organoids, organs-on-a-chip and other testing methods.
- Whether its approach produces safer candidates or fewer failures in clinical development.
Discussion spark: If animal tests are poor predictors of human toxicity, what evidence should regulators require before trusting AI predictions to guide drug development?
Sources and evidence
- Isomorphic Labs' Max Jaderberg: Rat Tests Are Only 30% Predictive of Human Toxicity – and That's the Case for AI Drug Discovery – BigGo Finance (12 November 2026, 13:08 UTC)
- DeepMind Researcher Ege Erdil: Frontier AI Labs Face Above-10% Bankruptcy Odds, and Memory Stocks Are the Better AI Bet – BigGo Finance (12 November 2026, 13:08 UTC)
Independent WittyWires tracker for public updates about Isomorphic Labs. Not affiliated with or endorsed by Isomorphic Labs; this is not an official account.