In February 2026, OpenAI initially launched Lockdown Mode, which was originally available only to a select group of ChatGPT users with high-security needs. It has now been extended to all individual users and self-service ChatGPT Business accounts. Lockdown Mode is turned off by default and can be activated by users as needed. Once enabled, certain features will be restricted, potentially affecting the user's daily experience. Specific limitations include: web browsing functionality will be limited to cached content only; image support may not display or search for images properly; image generation functionality will be limited, but users can upload images for AI to understand the content; Agent Mode and Deep Research functionality will be completely disabled; the Canvas networking feature will not allow generated code to access the network or external API endpoints; file download functionality will be restricted, preventing users from downloading files for data analysis, though manual file uploads for analysis will still be possible. While Lockdown Mode cannot completely prevent prompt injection attacks, it effectively cuts off data leakage pathways. Once enabled, ChatGPT will be unable to connect to external networks to transmit data, ensuring that even if prompts attempt to induce the AI model to leak data, the connection will be promptly severed. For users seeking higher security, enabling Lockdown Mode for handling sensitive data is a viable option. However, the best security practice remains refraining from sending sensitive data to AI models or using internally deployed AI models to process sensitive data. To enable Lockdown Mode, users can visit the ChatGPT security settings page (https://chatgpt.com/#settings/Security), navigate to ChatGPT Settings → Security → Advanced Security Settings, and activate Lockdown Mode.