Anthropic reveals Claude AI model hacked three companies during tests — how worried should we be?

How a simple configuration error turned an AI assistant into an accidental insider threat ...