AI tools are now making significant inroads into the biomedical field, and this time, Claude is stealing the spotlight.
On August 18, 2026, Anthropic announced that they had used their own Claude model to design a batch of proteins from scratch that had never existed in nature. [1]
They even used the designed proteins to spell out the company name "ANTHROPIC" in a dynamic demonstration, like this—
This brings to mind last weekend, when company founder Dario Amodei boldly claimed on X that with AI assistance, humanity could conquer most diseases within 5-10 years. [2]
Only a few days have passed—could the prophecy really be coming true so quickly?
To answer this question, we need to ask a few others: What exactly did Claude do? How impressive is this achievement? How far are we from actual drugs? Why are AI companies so focused on this area? Are AI-designed proteins safe? Is Amodei just blowing smoke?
So, how exactly did Claude pull this off? When it comes to proteins and AI, most people's first thought is Google's AlphaFold. Demis Hassabis and John Jumper from AlphaFold won the 2024 Nobel Prize in Chemistry for using AI to analyze and predict protein structures.
The other person who shared the Nobel Prize with them was David Baker, whose research direction was quite the opposite: designing proteins from scratch.
And Claude's work this time is similar to Baker's. The goal is creation, starting from zero, reverse-engineering a structure that doesn't exist in nature, which is a much higher level of difficulty.
Strictly speaking, what Claude designed should be called "protein binders," which function by using new proteins to bind to existing ones.
However, don't think of Claude as too magical. It's just a very powerful large language model, not an all-powerful Doraemon. In this protein design experiment, it acted more like a cyber-scientist, adept at leveraging various research tools to create an assembly line.