OpenAI Drops IH-Challenge Dataset to Harden AI Against Prompt Injection Attacks

OpenAI's new IH-Challenge training dataset improves LLM instruction hierarchy by up to 15%, strengthening defenses against prompt injec...