OpenAI Unveil lockdown mode to protect sensitive data from prompt injection

OpenAI We announced that the malicious chatbot command provides additional protection from prompt injection that will be hidden in web pages and other content sources.

In lockdown mode, in particular, the live web browsing (can access cached content only), the acquisition and display of images from the web (can generate images), detailed surveys, and agent modes are disabled.

According to the company, even if the lockdown mode is turned on,ChatGPT Prompt injections may still be vulnerable to prompt injections, for example, prompt injections may appear in cached web content or uploaded files that may affect response behavior and accuracy.

However, the goal is to reduce the possibility that sensitive data is shared in the process.

"Lockdown mode is not intended for everyone," says OpenAI. “This is designed for people and organizations that need more stringent protection from data leakage risks。 with instant induction handling sensitive data.

Live ChatGPT Business that you are deploying lockdown mode to your Business account and the target personal account.