OpenAI Lockdown Mode: ChatGPT Fights Prompt Injection

OpenAI has launched a new Lockdown Mode for ChatGPT. This optional security setting limits the AI's connections to the web and external services. It's a direct response to prompt injection attacks, where hidden malicious instructions try to make the AI leak sensitive data. When enabled, Lockdown Mode disables many of ChatGPT's network-connected capabilities. For example, live web browsing is replaced with cached content, and deep research and agent mode are turned off. ChatGPT cannot retrieve or display images from the web, and canvas-generated code cannot access the network. It also cannot download files for data analysis. This mode focuses on preventing sensitive data from being leaked. It is currently rolling out to ChatGPT Business accounts on the self-serve plan and to personal users. Administrators of managed workspaces can also assign this mode as a role-based permission. This new feature aims to give users more control over their data security when using ChatGPT.