Claude did not invent scorpion antivenom, experiment shows

Citrini, an investment research firm, published a blog post titled "We Wanted to Use Claude to Create a New Drug. Here's What Happened." The post described the team's limited biology background and its attempt to work from scratch; the author said "none of us could claim to know anything about AI as it related to biology." According to the account, Citrini first ran into Claude's biosecurity guardrails while researching drug candidates, then contacted people in its network who worked at AI labs and eventually found Anthropic willing to collaborate. Anthropic agreed to lift restrictions and give the team access to a more advanced, unreleased model on condition that the work be done at its offices, according to the post. Citrini said Anthropic also covered compute and token costs of just under $50,000.

The team chose scorpion antivenom, a costly treatment that is difficult to produce and often involves injecting venom into horses. Its goal was to use Claude to design proteins that could bind to toxins in scorpion venom. Citrini said it used a 35-page prompt and about 300 subagents to search for suitable proteins. Anthropic, in its own account, said Claude generated nearly 320,000 candidates and reduced them to 180 designs after discarding those that resembled known toxins or existing proteins. The process took 64 hours. Anthropic said the team worked "with Claude" to add 31 more designs, for 211 total.

Citrini sent the designs to two outside laboratories, Swiss biotech company Adaptyv Bio and U.S.-based Twist Biosciences, to test whether the proteins would bind to their intended targets. The tests cost about $150,000 more, according to the report. Seventeen proteins passed, collectively binding to four of the five scorpion toxins the team had hoped to target. Citrini said none of the proteins bound to more than one toxin.

Binding to a toxin is not the same as neutralizing it, and it does not establish that a protein can be a safe and effective treatment. Citrini acknowledged in its post that no experiment had tested whether the proteins actually stopped the toxin from working. Andreas Hougaard Laustsen-Kiel, a professor of antibody technologies at the Technical University of Denmark, told The Verge that finding a good binder against snake and scorpion toxins is relatively easy, while translating it into a medicine is the hard part. Melisa Benard Valle, an assistant professor, and Anne Ljungars, a senior researcher, both at the Technical University of Denmark, said in a joint email that the results were technically promising but that some of the excitement was premature. They described the findings as "far from a medicine and hyped" and said much remains unknown about the proteins' safety, stability, behavior in the body and potential to become viable treatments.

The experiment follows other Citrini projects that drew attention, including a fictional account of an AI-induced economic catastrophe that spread on social media and affected real-world markets, according to The Verge. For Anthropic, the collaboration comes as the company is pursuing drug development and life sciences and is weeks away from a highly anticipated IPO, according to the report. The Danish researchers said the most novel part of the experiment was the autonomous discovery process rather than the biological results, and that AI has substantial potential to accelerate discovery.