OpenAI has introduced Lockdown Mode, a security feature designed to reduce the risks associated with prompt injection attacks. By limiting access to certain capabilities, the company aims to better protect sensitive information from potential exposure through cached web content and uploaded files.
Enhanced Security Through Restrictive Access
Several features are restricted when Lockdown Mode is enabled. Live web browsing is unavailable, meaning users can access only cached content. Retrieval of images from external sources is also disabled, although image generation remains available. Additional capabilities, including Deep Research and Agent Mode, are limited as part of the security framework aimed at reducing data exposure risks.
Follow THE FUTURE on LinkedIn, Facebook, Instagram, X and Telegram
Targeted Rollout For High-Security Users
OpenAI has said that Lockdown Mode is not intended for all users. The feature is designed for individuals and organisations handling sensitive information that requires additional safeguards against data exfiltration risks. The rollout is taking place gradually across self-serve ChatGPT Business accounts and a limited number of eligible personal accounts.
Addressing Persistent Vulnerabilities
OpenAI has acknowledged that Lockdown Mode does not eliminate all prompt injection risks. Malicious instructions embedded in cached content or uploaded files may still influence model behaviour and responses. However, the company said the feature is designed to reduce the attack surface and strengthen protections for users working with sensitive data.
