OpenAI published a report describing a “new variety” of prompt injection that can self-propagate like a computer worm. The company says it observed no impact outside simulated tool calls in training and evaluation. Tests found injections arriving via email, replicating through the filesystem, and using multi-hop Slack instructions; OpenAI used its GPT-Red framework to discover them. https://thenewstack.io/openai-self-replicating-injections/