Anthropic's biology lab says AI agents discovered novel enzymes

Anthropic's recently formed biology lab reported that autonomous AI agents discovered novel enzymes with potential biotechnology applications, according to The New York Times. The company stressed that the results are preliminary and have not been peer-reviewed. Even so, the claim marks one of the most concrete examples yet of a frontier lab presenting agent-driven scientific discovery rather than benchmark performance.

The agents were described as operating with substantial autonomy across the discovery workflow: proposing candidates, running analyses and narrowing to promising enzymes. Autonomous research agents are becoming a crowded field. CMU and PhAI Labs recently released their open ScienceBuddy research agent on Hugging Face, and Google DeepMind has long pitched scientific discovery as AI's biggest payoff.

The timing is awkward and revealing. The discovery claim landed as CEO Dario Amodei joined Sam Altman in publicly calling for a safety slowdown, and as both were summoned before an Australian Senate AI inquiry on October 1. Anthropic is pitching autonomous agents as engines of scientific progress while acknowledging, alongside the industry, that agent containment is unsolved. Biology is also the domain where autonomy draws the most dual-use concern. Reuters reporting on Anthropic's IPO prospectus notes sweeping ambitions and surging costs, and a headline scientific result helps justify that spending.

Skeptics will want peer review, replication and wet-lab validation of enzyme function before treating this as discovery rather than candidate generation. Watch for a preprint, independent assessment by enzyme engineers, and details on biosecurity safeguards governing the lab's agents.