Research Firm Uses Anthropic Model to Design Scorpion-Venom Binding Proteins

Citrini Research, an investment research firm, published a report on Oct. 20, 2026, detailing a months-long experiment in AI-assisted protein design. The firm said it used Anthropic's internal research model with a $200,000 budget to design proteins intended to bind to scorpion venom, potentially providing a starting point for future antitoxin research.

According to the report, three of the five participants had no scientific background. In August 2026, the team submitted a roughly 35-page prompt to Claude. The model reportedly launched more than 300 sub-agents and used more than 300 GPUs in parallel overnight to design, score and screen proteins. The team then sent 211 designs to Twist Bioscience for synthesis; 17 met the binding criteria, exceeding the team's expectations.

The report cautioned that binding is not the same as treatment. The experiment did not test whether the proteins could block the toxin, nor did it demonstrate safety or efficacy in humans. Citrini described the work as an illustration of AI's potential to accelerate early-stage molecular design rather than the creation of a new drug.

Based on the experiment, the firm argued that as AI shortens the time from research ideas to experimental protocols, more candidate molecules will need validation in real laboratories, pushing the industry bottleneck toward testing capacity, experimental resources and execution. Citrini said it is bullish on contract research organizations (CROs) and companies providing lab resources, naming Charles River Laboratories, Revvity and Tecan Group as examples.