‘Not perfectly aligned’ with human values: Anthropic admits security failures behind AI hacking incidents

"The US startup behind the Claude chatbot has admitted a series of hacking incidents involving its models reflected a “failure of operational security” and said it has tightened its testing procedures."