Anthropic and Citrini Use Claude to Generate Binders for Five Scorpion Toxins
Researchers expanded a previous AI-driven protein design effort to target five scorpion neurotoxins, testing 211 molecular candidates in laboratory settings.
Key points
- Anthropic and Citrini Research used an internal Claude model to generate protein minibinders against five scorpion neurotoxins.
- The team tested 211 specific designs in the laboratory to evaluate their binding capabilities.
- This work expands on a previous August demonstration where Claude designed binders for a single toxin after receiving one prompt.
Anthropic and Citrini Research have used an internal version of the Claude artificial intelligence model to autonomously generate protein minibinders targeting five distinct scorpion neurotoxins. According to a report by Techmeme, the partnership produced 211 specific design candidates that were subsequently subjected to laboratory testing to verify their structural properties and binding efficacy.
The collaboration represents an expansion of earlier work published in August, when the same teams demonstrated that Claude could design protein minibinders after receiving a single text prompt. In that initial experiment, the AI system generated hundreds of candidate molecules, with laboratory tests confirming tight binding to a specific Mexican scorpion neurotoxin, as Threat Times reported on October 21, 2026.
Expanding the Target Scope
The current effort broadens the scope of the initial experiment by targeting five different scorpion neurotoxins rather than just one. Anthropic stated that the internal Claude model generated the protein structures autonomously, producing the 211 designs that were then sent for physical testing. This move demonstrates the model's ability to apply its protein design capabilities across multiple distinct biological targets without requiring entirely new foundational architectures.
By testing a larger variety of neurotoxins, the researchers aimed to validate the generalizability of the AI-driven design process. The source material confirms that the model was used to generate these specific minibinders, which are small protein structures designed to attach to specific target molecules. The laboratory phase focused on confirming whether these computationally generated designs could physically bind to the intended toxins in a controlled environment.
Research Context and Implications
This development follows the August disclosure where Claude successfully designed binders for a single toxin. The progression from one target to five suggests a systematic approach to testing the limits of AI-generated protein engineering. The involvement of Citrini Research, a laboratory specializing in protein engineering, was crucial for the physical validation of the computationally generated designs.
The partnership highlights the growing intersection between large language models and structural biology. While the initial August report focused on a single neurotoxin, the current work by Anthropic and Citrini Research, as reported by Techmeme, emphasizes the scalability of the method. The 211 designs tested represent a significant volume of experimental work, allowing researchers to assess the consistency of the AI's output across different biological challenges.
What should you do?
Security and technology professionals should monitor developments at Anthropic and Citrini Research for further disclosures on AI capabilities in biological domains. Organizations using AI for scientific research should review their protocols for validating computationally generated biological designs. Researchers should continue to assess the accuracy and safety implications of autonomous protein design tools.
Background: Large language models
Large language models are a specific type of neural network trained on text. Generative AI is the broader umbrella term for any system that creates new content, including images and code. Not all generative AI uses large language models, but most modern text generators do.
What to do and how to stay safe: Anthropic
- Review internal policies for the use of AI models in biological or chemical research to ensure appropriate validation steps.
- Monitor announcements from Anthropic regarding updates to Claude's capabilities in scientific domains.
- Verify the provenance of any AI-generated biological data used in experimental planning.
- Consult with domain experts when integrating AI outputs into physical laboratory workflows.
Step-by-step guide: Prompt Injection Warning Signs: How to Spot AI Manipulation
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
What did Anthropic and Citrini Research test in their latest experiment?
They tested 211 protein minibinder designs generated by an internal Claude model against five scorpion neurotoxins.
How does this differ from the August 2026 announcement?
The August announcement involved designing binders for a single Mexican scorpion neurotoxin, whereas this effort targets five different neurotoxins.
Did the AI system operate without human input?
The source material states the model autonomously generated the designs, but it does not specify the extent of human oversight or input during the process.
