OpenAI's Lockdown Mode hits millions: new security layer for ChatGPT

OpenAI is rolling out Lockdown Mode to millions of eligible users across Free, Plus, Pro, and self-serve ChatGPT Business plans. This optional setting limits features that connect to the web or external systems—including browsing, Deep Research, agent mode, file downloads, some image tools, and Canvas networking—to curb prompt injection attacks and data exfiltration. Prompt injection can hide malicious instructions in web content or files, tricking the chatbot into exposing sensitive data. The update gives IT teams a way to create a stricter boundary for sensitive ChatGPT work without blocking the entire platform. Memory, file uploads, and conversation sharing remain unchanged, so core functionality still works.

OpenAI explicitly states that Lockdown Mode reduces risk but does not guarantee complete security. Malicious instructions in cached web content or uploaded files may still affect behavior or accuracy. The tradeoff is usability: disabling research and automation features means organizations must carefully define when employees should enable Lockdown Mode, especially for confidential documents or regulated data. For IT leaders, this provides a clearer security layer, but safer AI use ultimately depends on prompt hygiene and internal policies. The expansion signals OpenAI’s growing focus on enterprise-grade controls as ChatGPT becomes more embedded in business workflows.