Health · Medical Research · 6 hrs ago
AI drug discovery faces its test in the lab
AI systems are helping researchers find biological clues that could lead to new medicines.
In September, nearly 1,000 versions of Anthropic’s Claude searched more than 200,000 enzymes and identified a CRISPR-like system in viruses that infect bacteria; scientists then confirmed it in the lab.
That lab confirmation is central to the approach: a model suggests an idea, researchers test it, and the results guide the next suggestion.
AI has already helped predict protein shapes, and one analysis found that 80% to 90% of AI-discovered drugs passed the first phase of clinical trials, compared with a historical average of 40% to 65%.
The promise is faster, cheaper drug discovery, which could make it worthwhile to pursue illnesses that attract little commercial investment.
Chagas disease affects an estimated 6 million to 7 million people, mostly in Latin America, and its only two medicines are more than 50 years old.
Whether AI can deliver on its ambitions depends on further lab tests, as researchers seek to discover treatments for many diseases within a decade.
Nearly 1,000 versions of Anthropic’s Claude searched more than 200,000 enzymes for 21 hours and identified a system resembling CRISPR.
Scientists then confirmed the finding in laboratory experiments.
AI drug discovery is being tested through a cycle in which models propose hypotheses and lab experiments check them.
The article says medicines discovered with AI are reaching patients, and reports higher-than-historical success rates in phase 1 trials.
Researchers hope cheaper drug discovery could make it worthwhile to develop treatments for diseases with limited commercial markets.
- Who
- Anthropic and other AI companies, alongside scientists testing AI-generated findings.
- What
- AI systems are being used to identify biological findings and help discover drugs, with laboratory experiments testing their predictions.
- When
- Anthropic’s enzyme search took place in September; the article was published on November 3, 2026.
- Where
- Not stated
- Why
- Laboratory tests show whether AI-generated hypotheses hold up in experiments, and cheaper drug discovery could make treatments for neglected diseases more viable.
This story does not have two clearly opposing sides.
OpenAI presented GPT-Rosalind, a model dedicated to biology.
Isomorphic Labs raised US$2.1 billion.
Nearly 1,000 versions of Claude searched more than 200,000 enzymes for 21 hours, and scientists confirmed a finding in the lab.
Kevin Weil launched Warp Sciences, aiming to build automated chemistry laboratories.
- Claude versions
- Nearly 1,000
- Enzymes searched
- More than 200,000
- Search duration
- 21 hours
- Isomorphic Labs funding
- US$2.1 billion
- Chagas disease cases
- 6 million to 7 million
- Phase 1 success rate for AI-discovered drugs
- 80% to 90%, compared with a historical average of 40% to 65%