AI-Designed Proteins Bind to Scorpion Toxins

A new research project demonstrates that AI sub-agents can cost-effectively synthesize proteins with specific binding properties.

Updated on Oct. 21, 2026 in Biotech

Isometric editorial illustration of a complex metallic protein structure on a clean laboratory table, representing AI-driven molecular design. — Citrini Research used AI sub-agents to synthesize proteins capable of binding to scorpion toxins, potentially accelerating future automated drug discovery workflows. AI Illustration. Upload story photo >

Live Poll

Do you trust artificial intelligence to accurately perform complex scientific research like drug discovery?

Citrini Research utilized an internal version of Anthropic's Claude to design 211 proteins, 17 of which successfully bound to scorpion toxins during laboratory testing. The research firm produced these results with a $200,000 budget, marking a move toward using AI-driven sub-agents for automated drug discovery.

Why it matters

This experiment highlights the growing utility of large language models in accelerating protein design and infrastructure development for AI-enabled virtual biology. It aligns with broader industry efforts to automate pharmaceutical research, following a recent collaboration between Anthropic and Novo Nordisk.

Using a 35-page prompt, the model spawned over 300 sub-agents to iterate designs, achieving low-nanomolar binding affinity for the Cn2-B1 and Cn2-B2 proteins. These results compare to manual laboratory discovery processes that typically require higher capital investment and longer timelines.

The players

Citrini Research

An investment firm and research house that identifies life sciences as a high-growth sector.

Anthropic

An AI safety and research company that develops the Claude large language model family.

Novo Nordisk

A global pharmaceutical company specializing in diabetes and obesity care that is currently integrating AI for drug discovery.

Twist Bioscience

A synthetic biology company providing DNA synthesis and genomics tools for laboratory research.

Adaptyv Bio

A biotechnology company that develops automated platforms for protein engineering and synthesis.

The details

To execute the design, Citrini Research utilized Claude to coordinate a network of more than 300 sub-agents, which functioned as autonomous units to iterate on structural protein models. The process involved direct prompting at Anthropic offices to circumvent standard biosecurity classifier restrictions that otherwise limit automated biological synthesis instructions. Synthesized proteins were subsequently validated through external laboratory testing by Twist Bioscience and Adaptyv Bio.

Early June: Citrini Research evaluated non-AI infrastructure sectors.
July: Anthropic unveiled Claude Science.
August: Citrini provided a 35-page prompt to Claude.
September: Novo Nordisk and Anthropic announced a collaboration.
Tuesday: Citrini Research published the project report.

The Tech Race

This project signals a clear acceleration in the move toward AI-native pharmaceutical pipelines, mirroring the trajectory of recent high-profile partnerships. The research underscores the competitive race to integrate LLMs into high-stakes biotechnology workflows previously dominated by manual bench science.

Investors and biotech firms can expect an increased supply of infrastructure designed to support AI virtual biologists, which may shorten discovery timelines for new drug candidates. This development is currently limited to research-stage applications and requires specialized synthetic biology services to implement.

The takeaway

The success of these designs demonstrates that LLMs can effectively coordinate complex biological design tasks with limited capital investment. Watch for upcoming announcements from companies providing infrastructure for AI-led protein engineering as this segment matures.