Investment Research Firm and Anthropic's Scorpion Venom Experiment Failed to Produce Medicine

Investment research firm Citrini Research collaborated with Anthropic to design antibodies against scorpion venom using an unreleased, advanced version of the Claude AI model. Conducted with a computing cost of approximately 50,000 dollars, the experiment saw Claude generate nearly 320,000 candidates, and external laboratory tests showed some proteins bound to targets. However, experts emphasized that these results are far from being a real medicine and that the process covers only the initial phase.

Citrini Research contacted Anthropic to see how artificial intelligence could impact drug development processes. Anthropic eased biosecurity restrictions to provide the team with a more advanced, unreleased Claude model at the company's offices and covered approximately 50,000 dollars in computing costs. Working with a prompt set of about 35 pages, nearly 300 sub-agents were tasked with designing proteins capable of binding to toxins in scorpion venom.

According to Anthropic's data, Claude generated approximately 320,000 candidates within 64 hours, and after eliminating those resembling known toxins, 180 designs were identified. With 31 designs added through the partnership of the team and Claude, the total number reached 211. Citrini paid approximately 150,000 dollars to Swiss laboratory Adaptyv Bio and US laboratory Twist Biosciences to test these designs. In the tests, 17 proteins managed to bind to four of the five targeted scorpion toxins, but no protein bound to more than one toxin.

Evaluation and Expert Opinions

Experts stated that the binding of proteins to a toxin does not mean it can be neutralized or become a safe therapeutic tool. Andreas Hougaard Laustsen-Kiel, Melisa Benard Valle, and Anne Ljungars from the Technical University noted that the results are technically promising but very far from being a medicine and are exaggerated. Scientists stated that the real innovation was the autonomous discovery process rather than the biological results, and that turning molecules into an effective medicine poses a much greater challenge.