OpenAI Launches Lockdown Mode for ChatGPT Security

OpenAI is enhancing security with its new optional Lockdown Mode for ChatGPT, designed to combat the growing threat of prompt injection attacks. These attacks aim to mislead AI models by inserting malicious instructions into conversations, potentially leading to data exfiltration.

Lockdown Mode operates by limiting outbound network requests. This significantly boosts protection against malicious prompts. However, enabling this mode comes with notable trade-offs that impact functionality. Users will find web browsing capabilities restricted to cached content, limiting access to fresh web pages.

Furthermore, ChatGPT's ability to display or retrieve images from the web will be disabled in Lockdown Mode. File downloads for data analysis and Canvas-generated code access to the network will also be unavailable. Advanced features like Deep Research, Agent Mode, and Developer Mode will be disabled.

This security setting is available for all account types but is specifically recommended for users and organizations handling sensitive data. OpenAI indicates that while Lockdown Mode substantially reduces risks, it does not guarantee complete protection against all novel attack techniques.