OpenAI Introduces Lockdown Mode and Elevated Risk Labels in ChatGPT

New security layer blocks prompt injection and data exfiltration in agent mode — sets a bar for production AI safety.