Anthropic says it blocked AI bioweapons research bids

Anthropic published a 154-page report Thursday cataloging how people tried to misuse its Claude chatbot. The most serious finding: five cases where scientists used Claude for research that could help build biological weapons. In one, someone asked Claude to help write a grant application. The topic: a virus's ability to spread and dodge immune systems. Anthropic's safety systems caught it and blocked the account.

Anthropic can't prove these scientists meant to build a weapon. It says so plainly in the report. But it blocked and banned them anyway. The potential harm was too serious to risk being wrong. The same report flagged other misuse too. Iranian accounts used Claude to track dissidents on social media. Separately, Chinese AI labs ran millions of hidden queries through Claude to help train their own competing models. Anthropic chose to publish all of this. Not every AI lab does.

Keep reading — what to watch

Will other AI companies start publishing reports like this? Right now, Anthropic is largely doing this alone. That means the public only sees what one company catches and shares. If a scientist tries the same request on a rival chatbot with weaker safeguards, nobody may ever know.