OpenAI has announced a new feature for ChatGPT called Lockdown Mode, designed to combat prompt injection attacks. This feature is targeted at users and organizations handling sensitive data, with the goal of reducing the risk of data leakage.
Disable part of the online connectivity
After enabling, ChatGPT will no longer be able to browse the web in real time and will only access cached content. Web image retrieval and display will be disabled, but image generation will still be available.
Deep research and agency mode are limited
OpenAI will also disable Deep Research and Agent Mode in this mode. The company stated that this reduces interference with model behavior when malicious instructions are embedded in web pages, files, and other content.
Risk cannot be completely eliminated.
OpenAI also cautions that even with Lockdown Mode enabled, ChatGPT may still be vulnerable to prompt injection. For example, malicious instructions could still be embedded in cached web content or uploaded files, potentially affecting the accuracy of responses.
Additional information: This feature is gradually being rolled out to ChatGPT Business self-service accounts and eligible individual accounts.