🚨 NEW: OpenAI says ChatGPT Atlas is being hardened against prompt injection attacks using automated, reinforcement learning–based red teaming to stop exploits before they reach users. https://t.co/tIaIyRLnfR ...