Citrini Research attempted to generate a functional scorpion antivenom using an unreleased version of Anthropic’s Claude model. The experiment aimed to test whether large language models could design a new drug entirely from scratch. The AI produced several promising starting points for chemical compounds. Experts told The Verge that the output fell far short of a viable medication.
The trial highlights a significant gap between theoretical AI capabilities and practical pharmaceutical requirements. Drug development demands precise biological knowledge that current models lack. Citrini admitted its team possessed no specific expertise in AI or biology. Without this foundation, the model could not verify its own suggestions for safety or efficacy. The result suggests that generative tools cannot yet replace human scientists in creating life-saving treatments.
* The model generated initial compound structures but failed to complete a full drug pipeline.
* Citrini acknowledged a lack of internal expertise in both AI and biological sciences.
* Experts confirmed the output was not close to an actual usable medicine.
