Can AI Create New Drugs? Results of a Recent Experiment
AI's ambitious attempt in pharmaceuticals
Exaggerated expectations around artificial intelligence (AI) capabilities have been mounting recently. This time, investment research firm Citrini Research set out to test whether a Claude model could create a pharmaceutical drug from scratch. They partnered with Anthropic, which provided an unreleased, enhanced AI model along with roughly $50,000 worth of compute resources for the experiment.
The goal of the experiment was to develop an antidote for scorpion venom. This process is traditionally complex and expensive, often requiring the use of animals. The team tasked Claude with designing proteins capable of binding to and neutralising toxins found in scorpion venom.
What actually happened?
After 64 hours of operation, Claude generated over 320,000 candidate proteins, which were filtered down to 211 options. These designs were sent to independent biotechnology companies for laboratory testing, costing an additional $150,000. Ultimately, only 17 proteins demonstrated the ability to bind to scorpion toxins. However, none were capable of working against multiple toxins simultaneously.
Although some proteins bound to toxins, experts unanimously point out that this does not equal creating a drug. Toxin binding is merely the simplest stage of the process. According to professors of biological sciences, it remains unknown how safe, stable, or effective these proteins would be in a living organism. "These results are technically interesting, but there is still a very long way to go before a real medication is created," specialists in the field noted.
What does this mean for the tech sector?
Following the experiment, Citrini declared the process a success, though purely in terms of investment potential. The firm acknowledged that its goal was not to invent a new drug, but to demonstrate AI's commercial capabilities. Scientists, meanwhile, warn the public that such projects are often steeped in hype and do not reflect genuine scientific breakthroughs.
Ultimately, this case highlights the divide between data processing and real-world medicine. Although AI can accelerate molecule discovery, developing a drug requires lengthy clinical trials, safety testing, and proof of biological stability—things software code alone cannot yet provide. For readers in Georgia following AI developments, this serves as a reminder that marketing claims from tech companies should always be evaluated through a critical scientific lens.