Editorial illustration for Anthropic’s Claude Generates Over 318,000 Candidate Antivenom Ideas
Claude Generates 318K Antivenom Candidates in 64 Hours
Anthropic’s Claude Generates Over 318,000 Candidate Antivenom Ideas
Five people with a hedge fund research background and no drug development experience just spent 64 hours trying to design a protein that neutralizes scorpion venom. Anthropic published the results on October 20, and the numbers are strange enough to stop on: 318,484 candidate binders generated, 211 sent off for lab testing, 17 that came back qualifying as hits against four of five toxin targets.
The team came from Citrini Research, an investment firm, not a biotech lab. Three of the five staffers had no science background at all. They worked with roughly $200,000 and an internal version of Claude, after biosecurity restrictions forced them out of their own offices and into Anthropic's, where staff helped them build out a prompt stretching about 35 pages.
What they were chasing isn't a finished antivenom. It's a set of proteins that bind to scorpion toxins in a dish, tested by two outside labs, Twist Bioscience and Adaptyv Bio. Whether any of those 17 binders can actually stop venom from hurting a living cell, let alone an animal or a person, is a separate question entirely.
The key is who held the keyboard. Staffers at an investment newsletter with little drug development experience brought an idea through computer design and into physical testing. That suggests AI could broaden who starts research projects by helping them coordinate specialist tools and turn a question into candidates for experiments.
Why this matters
The numbers here are the story, and they cut both ways. Generating 318,484 candidates in 64 hours is a genuine demonstration of throughput: no team of five researchers, however capable, screens that many protein designs by hand. But throughput isn't validation.
Of 211 designs that made it to the lab, only 17 qualified as binders, and binding to a toxin is not the same as neutralizing venom in a living organism. That gap between "computationally promising" and "biologically proven" is where this experiment actually lives, and it's worth sitting with before anyone calls this an antivenom breakthrough. For developers and founders watching Anthropic push Claude into wet-lab science, the real signal is the 35-page prompt and the on-site Anthropic staff support needed to make this work.
That's not a product yet, it's a research collaboration with heavy scaffolding. The interesting question for our readers isn't whether Claude can propose candidates at scale. It's whether that scaffolding can be stripped down enough for ordinary labs to replicate without flying a team into Anthropic's offices.
Related Reading
Browse all Anthropic news →
Business & Startups Nordic pilot adds Gemini for Education, NotebookLM to boost AI literacy
Business & Startups Kling launches Video O1, all-in-one model with MVL bridge using transformer
Business & Startups DeepSeek Seeks More Capital Weeks After USD 7B Funding Round
Related Topic Claude gains shared context in Excel, PowerPoint; Microsoft adds Copilot Cowork
AI Tools & Apps
Related Topic Anthropic: Hackers Stole Claude Tokens Via Compromised Session Key
LLMs & Generative AI
From Archives Accounting Professor Notes Redaction of Data Center Lease Details Is Common
From Archives OpenAI turned coding tools into dots, an always-on personal assistant
From Archives Claude Code vs Cursor: Field Test Shows No Single Winner in Real Use
LLMs & Generative AI
From Archives Anthropic Burns Gas to Feed AI Boom Amid Intense Competition for Power
Market Trends
Common Questions Answered
How many candidate antivenom proteins did Claude generate for the scorpion venom project?
Claude generated 318,484 candidate binders in just 64 hours as part of the Anthropic research project. Of these candidates, 211 were sent to laboratory testing, and 17 ultimately qualified as hits against four of the five toxin targets being studied.
What was unusual about the team that conducted the scorpion antivenom research?
The team came from Citrini Research, an investment firm, rather than a traditional biotech laboratory, and had a hedge fund research background with no prior drug development experience. Despite their lack of specialized training, they were able to use Claude to coordinate specialist tools and generate viable protein candidates for experimental testing.
What is the significance of AI broadening who can start research projects according to this study?
The study demonstrates that AI tools like Claude can help non-specialists with limited drug development experience design and execute complex research projects by coordinating specialist tools and converting research questions into testable candidates. This suggests AI could democratize research initiation beyond traditional biotech institutions and trained researchers.
What is the gap between Claude's computational results and actual biological validation in this antivenom project?
While Claude generated impressive throughput with 318,484 candidates in 64 hours, only 17 of the 211 lab-tested designs qualified as binders, and binding to a toxin is fundamentally different from neutralizing venom in a living organism. This highlights the distinction between computationally promising results and biologically proven efficacy in drug development.
Further Reading
- Claude did not invent scorpion antivenom - The Verge
- How Claude designed protein binders against scorpion neurotoxins - Anthropic
- Wall Street researchers just used Claude to design a drug. The experiment has spawned a new investment thesis. - MarketWatch
- Claude's 64-hour experiment finds 17 binders for scorpion toxins - Interesting Engineering
- Claude Science tackles deadly scorpion stings by designing protein binders from scratch - The Decoder
