Bottom line: Lockdown Mode disables multiple functions with network access but offers no absolute protection against prompt injection attacks.
OpenAI has introduced an optional security feature called Lockdown Mode for ChatGPT to reduce the risk of data exfiltration through prompt injection attacks. The feature is being rolled out incrementally to users of all subscription tiers – Free, Plus, Pro, and Business plans.
Lockdown Mode is an optional advanced security setting that restricts tools and functions in ChatGPT that can connect to the web or external services. The primary goal is to block the digital pathways through which data exfiltration to attacker-controlled servers could occur. To do so, the mode disables or restricts the following functions: live web browsing (only cached content permitted), image support, Deep Research, agent mode, Canvas networking, and file downloads within data analysis. Lockdown Mode cannot be active simultaneously with Developer Mode; activating one automatically deactivates the other.