An investment research firm wanted to prove that AI could transform drug development. It did not come close to making a drug. Citrini Research teamed up with Anthropic to see whether an unreleased version of Claude could help design a scorpion antivenom. The project produced some promising starting points, but experts tell The Verge that nothing in it resembles an actual medicine.
Citrini is best known for a widely circulated piece of fiction that imagined an AI-induced economic catastrophe. The story gained traction on social media and rattled real-world markets. The firm later wrote a blog post titled “We Wanted to Use Claude to Create a New Drug. Here’s What Happened.” In it, Citrini admitted that “none of us could claim to know anything about AI as it related to biology.”
The collaboration began with a roadblock. Citrini ran into Claude’s biosecurity guardrails while trying to get the model to research drug candidates. The post’s author said they “spam called anyone in my contact list who worked at an AI lab” until they found a willing partner. Anthropic agreed to drop the restrictions and give the team access to a more advanced, unreleased model, on one condition: they had to work from its offices. Anthropic also agreed to cover the compute and token bill, which came to just under $50,000.
The team settled on scorpion antivenom, a costly treatment that can be difficult to produce and often involves injecting venom into horses. The goal was to use Claude to design proteins that could bind to the toxins in scorpion venom. Citrini armed the model with a 35-page prompt and deployed some 300 subagents to hunt for suitable proteins.
Anthropic published its own, more restrained account of the work. According to the company, Claude generated nearly 320,000 candidates. The team discarded those that resembled known toxins or existing proteins, leaving 180 designs. Working “with Claude,” the team added 31 more, bringing the total to 211. The entire process took 64 hours.
How did researchers test the Claude-designed proteins?
Citrini then sent the designs to two outside laboratories: Swiss biotech company Adaptyv Bio and US-based Twist Biosciences. Testing whether the proteins could bind their intended targets cost another $150,000, give or take. It is unclear why Anthropic did not run those tests in-house at its biology wet lab in California. Seventeen proteins passed, collectively binding to four of the five scorpion toxins the team had targeted. None of them bound to more than one toxin, Citrini said.
Does binding to a toxin make an antivenom work?
Even a successful binding result would fall well short of an antivenom. Binding to a toxin does not mean a protein can neutralize it, much less serve as a safe and effective treatment. Citrini acknowledges as much, writing that “no experiment has yet tested whether they actually stopped the toxin working.”
Researchers who reviewed the work say the team accomplished the easiest part. “It is easy-ish to discover a good binder against snake and scorpion toxins,” said Andreas Hougaard Laustsen-Kiel, a professor of antibody technologies at the Technical University of Denmark. “The hard part is to translate it into a medicine.”
Is AI-designed scorpion antivenom a finished medicine?
Melisa Benard Valle, an assistant professor, and Anne Ljungars, a senior researcher, both at the Technical University of Denmark, were blunter in a joint email. “The results are technically promising, but some of the excitement is premature,” they wrote, calling the findings “far from a medicine and hyped.” Much remains unknown about the Claude-designed proteins, they said, including their safety, stability and how they behave in the body. In their view, the most novel part of the project was the “autonomous discovery process rather than the biological results.”
Andreas Bender, a professor of machine learning in medicine at Khalifa University in Abu Dhabi and a visiting professor at the University of Cambridge, said the underlying approach is not new. Existing methods such as high-throughput screening “are often able to find a multitude of hits at manageable cost,” even without AI. By his account, designing proteins that bind to targets is “kind of” a solved problem. The harder task is turning those binders into treatments that work in living systems. “It’s exactly this translation that lacks,” he said.
Citrini still counted the experiment a win of sorts. The firm argued that it offers a glimpse of the investment opportunities emerging from AI-enabled drug discovery. When critics questioned why the firm did not take the research further, Citrini accused them of “nitpick[ing]” and wrote, “Not to beat a dead horse, but we are an investment research firm - not a biotech startup.”
The firm’s investment case is reserved mostly for paying subscribers. Bloomberg reports that Citrini highlighted companies involved in the testing side of drug development, including Charles River Laboratories, which supplies primates for animal research. In a video about the experiment, Citrini described the AI portion as “free,” even though Anthropic covered the compute costs. The Verge noted that Anthropic, weeks away from a highly anticipated IPO, has its own reasons to promote AI’s potential in drug development and the life sciences.
This article was written with the assistance of AI.
News Factory APP - agentic news to boost your SEO & AEO.