THE CRUNCH

Anthropic has documented users finding ways around Claude's safeguards for bioweapons research. The company's threat intelligence report details eight months of abuse, including attempts to generate instructions for dangerous biology.

The Decoder reports that the threat report covers a wide range of misuse, including the creation of missile software, autonomous kamikaze drones, and nationwide surveillance systems. The report also highlights that Chinese AI labs, such as Alibaba's Qwen team, DeepSeek, and Moonshot AI, relayed requests en masse or extracted training data, with Qwen alone accounting for more than 151 million exchanges.

WHAT HAPPENS NEXT

Anthropic will likely need to update its safety controls and monitoring systems to address these new bypass techniques and the scale of external data extraction.