Editorial illustration for Claude Science Designs Scorpion Venom Protein Binders After 24-Hour Run
Claude AI Designs Scorpion Venom Protein Binders
Scorpion stings kill more than 3,250 people a year, mostly in regions with limited access to antivenom. Anthropic set its Claude model loose on a piece of that problem: designing new protein binders that could neutralize the neurotoxins responsible. Protein binders are small molecules built to grab onto one specific target, in this case toxins pulled from four scorpion species.
Working inside Claude Science, Anthropic's research environment, Claude got a single prompt and then ran on its own for 24 hours straight. In that stretch it generated 318,484 candidate designs aimed at five different toxins. Researchers stepped back in afterward to guide the next phase, narrowing the field before anything reached a lab bench.
This wasn't Claude's first attempt at this kind of work. An earlier study had it designing mini-binders against 14 of 15 target proteins, hitting success rates of 22 to 35 percent, well above the usual 10 to 15 percent researchers see with conventional methods. The scorpion venom project pushed that approach further, from raw generation through selection and into actual wet-lab testing.
Claude picked 180 candidates, and the team added 31 more. Of the 211 designs tested in the lab, 17 met the criteria for a binder. The two strongest bound extremely tightly to Cn2, a toxin from the Mexican scorpion.
Why this matters
A hit rate of 17 out of 211 lab-tested designs, drawn from a pool of 318,484 candidates generated largely without human input, is the number worth sitting with. That's not a cure for scorpion envenomation, which still kills more than 3,250 people a year, mostly in regions with little access to antivenom. But it's a real signal about where AI-run science is heading: a model that works unsupervised for a full day, then takes direction instead of needing constant hand-holding.
For researchers, the interesting part isn't the two tightest-binding candidates, it's the workflow. Anthropic's team still had to pick 180 of Claude's designs and add 31 of their own, so this is augmentation, not replacement. Founders building biotech tools should note the funnel: generation is cheap and fast now, wet-lab validation is still the bottleneck, and that gap is where products will get built.
For anyone skeptical of AI hype in science, the honest number here is the ratio, not the headline. Worth watching whether that 8 percent success rate holds up as Claude tackles targets with less data behind them.
Related Reading
Browse all Anthropic news →
LLMs & Generative AI ChatGPT's 'Nerdy' tweak rewards goblin metaphors in answers, study finds
LLMs & Generative AI Google to End Free Access to Gemini AI for Non-Subscribers
LLMs & Generative AI AI Engineers Face Rising Costs, Need New Strategies for Efficiency
Related Topic Claude gains shared context in Excel, PowerPoint; Microsoft adds Copilot Cowork
AI Tools & Apps
Related Topic Trump cracks down on Anthropic after Amazon tip; staff largely foreign
Open Source
From Archives Continuum-LM Reuses Dynamical Operators in Continuous-Depth AI4Science Models
From Archives Order effect aligns Qwen, Llama, Gemma transformer paths 1.60–2.24x
From Archives Anthropic pitches AI safety bill to block models if 50+ could be killed
Policy & Regulation
From Archives Anthropic co-founder says singularity underway amid AI security focus
Business & Startups
Common Questions Answered
How many protein binder designs did Claude generate and what was the success rate in lab testing?
Claude generated 318,484 candidate designs largely without human input, from which 211 were selected for lab testing. Of those 211 designs tested, 17 met the criteria for effective binders, representing an 8% success rate that researchers consider a significant achievement for AI-driven protein design.
What specific scorpion toxin did the strongest protein binders target?
The two strongest protein binders designed by Claude bound extremely tightly to Cn2, a neurotoxin extracted from the Mexican scorpion. This represents a major breakthrough in targeting one of the most dangerous scorpion species responsible for human deaths.
How long did Claude run unsupervised to design these protein binders?
Claude ran autonomously for a full 24-hour period within Anthropic's Claude Science research environment on a single initial prompt. This extended unsupervised operation demonstrates a significant advancement in AI-driven scientific research that requires minimal human intervention or constant direction.
Why is this protein binder research important for addressing scorpion sting deaths?
Scorpion stings kill more than 3,250 people annually, particularly in regions with limited access to antivenom. Developing new protein binders that can neutralize scorpion neurotoxins could provide life-saving treatment options in areas where traditional antivenom is unavailable or inaccessible.
What makes this AI-designed protein binder project different from previous antivenom development approaches?
This project demonstrates AI working unsupervised for extended periods and then taking direction rather than requiring constant human guidance throughout the design process. The ability to generate and test hundreds of thousands of candidates efficiently represents a shift in how AI can accelerate scientific discovery compared to traditional manual protein design methods.
Further Reading
- How Claude designed protein binders against scorpion neurotoxins - Anthropic
- Claude-assisted protein designs reportedly yield 17 scorpion toxin binders - Digg
- Claude Designed 17 Scorpion Antivenom Binders in 64 Hours - CellCog
- Anthropic’s Claude Designed 17 Working Protein Binders Against Scorpion Venom Toxins - OfficeChai
- Anthropic says Claude designed binders for scorpion neurotoxins - Blockchain.News
