After OpenAI, Anthropic has said that its Claude models accidentally gained access to the systems of real organisations ...