Anthropic Has Quietly Been Running an Actual Biology Lab

Anthropic, the company that has made AI safety its founding principle and spent considerable time warning the public about the existential risks of the technology it builds, is now operating a physical biology laboratory. This is not a metaphor. According to TechCrunch, Anthropic researchers are conducting real biological experiments, using AI to accelerate scientific discovery in the life sciences. The detail is tucked inside a broader story about Anthropic's expanding research footprint, but it deserves more attention than it is getting.

The context here is important. Anthropic co-founder and CEO Dario Amodei has said publicly that he believes AI could compress decades of scientific progress into a few years, and that medicine and biology are among the fields most likely to be transformed. The company has published research suggesting its Claude models show promise in scientific reasoning tasks. A biology lab is, in that framing, the logical next step: a place to test whether AI-assisted research actually produces useful results, not just impressive benchmark scores.

The Dual-Use Tension

But there is a tension here that the company itself has acknowledged. Anthropic's safety researchers have published work on biosecurity risks from AI, including the concern that sufficiently capable AI systems could lower the barrier to developing biological weapons by making specialised knowledge more accessible. The company has instituted usage policies specifically designed to prevent Claude from providing detailed guidance on dangerous pathogens.

Running a biology lab while simultaneously warning about AI-enabled biosecurity risks is not inherently contradictory. Research institutions have always had to navigate dual-use concerns, the same knowledge that helps you develop a vaccine can, in theory, inform harm. But it does raise questions about governance and oversight that are worth asking clearly. What experiments are being conducted? What safety protocols are in place? Who has visibility into the research programme?

These are not accusatory questions. Anthropic is not a rogue actor. It is a company that publishes more about its internal safety thinking than almost anyone else in the frontier AI space. But operating a biology lab is a meaningful escalation of scope, and the public interest in transparency around that seems legitimate.

Why AI-Assisted Biology Is a Big Deal

Setting aside the dual-use concerns for a moment, the potential upside of AI-accelerated biology research is significant enough to take seriously. Drug discovery is extraordinarily slow and expensive. The average time from target identification to approved drug is around twelve years, and failure rates at every stage of clinical development are high. AI tools that can propose better candidate molecules, predict protein interactions, or identify existing compounds with new applications could genuinely save lives and reduce costs at scale.

Anthropicis not the only organisation pursuing this. DeepMind's AlphaFold changed how the field thinks about protein structure prediction. Isomorphic Labs, also a DeepMind spinout, is doing AI-first drug discovery. Recursion Pharmaceuticals has been running AI-assisted experiments for years. What is different about Anthropic's approach is that it is being pursued by a company whose primary product is a general-purpose language model, and whose stated reason for existing is to ensure that advanced AI does not cause catastrophic harm.

The combination of those two things, genuine scientific ambition and genuine safety concern, is either a strength or a complication, depending on your perspective. The optimistic reading is that Anthropic is better placed than most to pursue this research responsibly. The sceptical reading is that running a biology lab is exactly the kind of capability expansion that makes the already-difficult job of AI governance harder.

What Anthropic Should Do Next

Publish. The company's track record on transparency is better than most of its peers, and extending that to its biology research programme would go a long way toward addressing legitimate questions about oversight. What experiments are in scope? What is the governance structure for the lab? How does biosecurity review work internally? These are answerable questions, and Anthropic is more capable than almost any organisation of giving good answers. The field, and the public, would benefit from them.