ChatGPT gets a lockdown mode — but developers still own the design
Prompt-Injection Defense (Developer View)OpenAI launched Lockdown Mode in ChatGPT: a high-security mode that disables live browsing, web images, deep research, and agent mode to curb prompt injection.
Until last month ChatGPT had no native mode to restrict attacks from external content — developers stacked guardrails and privilege separation entirely on their own.
Lockdown Mode shrinks the live-web and agent attack surface, but cached pages and uploaded files remain vectors — privilege isolation is still your job. Minimal for casual users.