Anthropic performing prompt injection on its users

Analysis of an unusual interaction between a user and the Claude AI model, where the model responds to a prompt injection attempt with a warning message, leading to speculation about the nature of the interaction.