Anthropic Says Its A.I. Systems Broke into Computers at 3 Organizations

Anthropic researchers found that their artificial intelligence models were able to successfully navigate and manipulate computer systems during security testing. By executing sequences of commands to gather information and perform tasks, the systems demonstrated an ability to interact with external software environments. This development highlights the potential for autonomous agents to misuse digital tools if not properly restricted. These findings underscore an emerging security challenge as AI systems gain the capability to perform complex, multi-step actions across various software platforms.