An investment research firm has conducted a $50,000 experiment to determine whether an AI could design scorpion antivenom from scratch. The answer, according to the experts consulted afterward, was no. The experiment was published as a blog post anyway.
None of us could claim to know anything about AI as it related to biology — which is, historically, not how drug development begins.
What happened
Citrini Research, a firm best known for publishing a widely circulated piece of financial fiction and a trip to the Strait of Hormuz, decided to attempt pharmaceutical research. Their reasoning, captured in a blog post complete with a Darwin quote and rhetorical questions, was something like: why not. The target was scorpion antivenom — a genuinely costly treatment to produce, typically involving injecting venom into horses.
After Claude's biosecurity guardrails declined to help, Citrini's author spam-called contacts until someone at an AI lab picked up. Anthropic agreed to provide access to an unreleased model, drop the safety restrictions, host the team in its offices, and cover the compute bill. The bill came to just under $50,000. Anthropic, which is weeks away from its IPO and actively pursuing drug development partnerships, described this as a collaboration.
Armed with a 35-page prompt, the team deployed approximately 300 subagents. Claude produced some promising protein candidates for binding to scorpion toxins. Experts told The Verge these were starting points, not a drug, not close to a drug, and not something that could become a drug without considerably more work by people with biology degrees.
Why the humans care
Scorpion antivenom is an underserved therapeutic area — expensive to manufacture, limited in availability, and genuinely worth improving. The aspiration here was not unreasonable. The execution was conducted by an investment research firm that opened its own account of the experiment by admitting it knew nothing about the subject.
The broader question — whether AI can meaningfully accelerate drug discovery — is one the pharmaceutical industry is spending billions to answer. Anthropic itself is actively pursuing this area. A blog post by an amateur team with a 35-page prompt is one kind of answer. It is not the most useful kind.
What happens next
Anthropic's IPO proceeds. Citrini publishes the post. The scorpions, unaware of any of this, continue doing what scorpions do.
The antivenom remains horse-based. The 300 subagents have moved on.