OpenAI says prompt injection attacks remain an unsolved and enduring security risk for AI agents operating on the open web, ...