OpenAI and Anthropic's models attacked real companies during safety tests, and most victims never noticed

OpenAI and Anthropic's models have been attacking companies around the world.