OpenAI has rolled out GPT‑Rosalind, a biology‑tuned large‑language model designed to flag dubious drug targets and exhibit skeptical reasoning. The model claims to handle multi‑step scientific queries, though its performance on expert benchmarks remains unverified.

Developed to curb the overenthusiastic tone of earlier LLMs, GPT‑Rosalind limits access to U.S. entities under a trusted‑access framework. A public Life Sciences Research Plugin will follow, but the company warns that the model could generate harmful outputs if misused.

While other firms offer generic science agents, GPT‑Rosalind’s narrow focus on biology marks a notable shift. Its true value will hinge on real‑world testing and the balance between insightful connections and potential hallucinations. The release signals a cautious push toward specialized AI in drug discovery.