Anthropic says it uncovered and stopped several instances where its Claude models were used in scientific research that could potentially feed into the development of biological weapons. The company detailed these cases in a new report focused on AI misuse. Anthropic made it clear they have no evidence the researchers actually intended to build weapons—the issue centers on the dual-use nature of this type of research.

What Anthropic's report reveals

The report lays out specific cases where Claude was tapped for biology-related projects, with results that could have dangerous applications down the line. Anthropic stresses that these are risks tied to dual-use scientific methods, not proof of malicious intent.

How the company responded

Anthropic says it blocked these attempts and included them in its report as examples of how AI could potentially be abused. The company is pushing for tighter controls and oversight on how language models are used in sensitive fields, aiming to keep neutral research from turning into something risky.

The dual-use dilemma

This whole episode shows that large language models can speed up scientific work—but they also need extra guardrails in high-bio-risk areas. When AI and biology overlap, the big question is how to support legitimate research without opening the door to dangerous uses.