Anthropic and Citrini Research Experiment Fails to Develop Functional Scorpion Antivenom via Claude AI
An experimental collaboration between investment research firm Citrini Research and AI lab Anthropic aimed to demonstrate the potential of large language models in drug discovery. Using an unreleased version of Claude, the team attempted to design proteins capable of binding to scorpion toxins to create a new antivenom. Despite utilizing 300 subagents and generating over 320,000 candidates, the process yielded only limited success. The experiment resulted in 17 proteins that successfully bound to specific toxins when tested by external laboratories Adaptyv Bio and Twist Biosciences. However, experts noted that binding does not equate to neutralization, meaning the project fell far short of creating a viable drug. The high-cost endeavor, which included $50,000 in compute costs and $150,000 in laboratory testing, serves as a cautionary tale regarding the current limitations of AI in complex biological engineering.