OpenAI Launches Lockdown Mode
To Defend ChatGPT Users Against Prompt Injection Attacks
News Desk
07 June 2026, 10:48
OpenAI has introduced a new security feature called Lockdown Mode, designed to provide stronger protection against prompt injection attacks that could expose sensitive information during interactions with ChatGPT.
Prompt injection attacks occur when malicious instructions are hidden inside webpages, documents, or other content sources accessed by an AI system. These hidden prompts can influence the chatbot’s behavior and potentially cause it to reveal information or perform unintended actions.
To reduce those risks, Lockdown Mode restricts several ChatGPT capabilities. When enabled, users will no longer have access to live web browsing, meaning the chatbot can only use cached web content. The feature also disables web image retrieval and display, while image generation remains available. In addition, both Deep Research and Agent Mode are turned off.
Despite the added safeguards, OpenAI acknowledged that Lockdown Mode does not eliminate prompt injection risks entirely. The company noted that malicious instructions could still exist in cached web content or uploaded files and may continue to affect the accuracy or behavior of responses.
According to OpenAI, the feature is intended primarily for individuals and organizations that regularly work with sensitive information and require stronger protection against data exfiltration attempts linked to prompt injection attacks.
The company emphasized that Lockdown Mode is not designed for all users, but rather for security-conscious environments where protecting confidential data is a priority.
OpenAI said the feature is currently being rolled out to self-service ChatGPT Business accounts as well as eligible personal account holders.
