ChatGPT's New Incident Mode: Prevent Data Theft

OpenAI has launched a new "Incident Mode" for ChatGPT to help prevent data theft. This optional security setting is available to all logged-in users and covers different account types and workspaces. Here's the thing: When activated, ChatGPT will restrict its real-time internet access, in-depth research, and agent modes. The goal is to build a firewall against prompt injection attacks by limiting the model's ability to initiate external network requests. In this mode, web browsing will be strictly limited to cached content, which might mean incomplete or outdated search results. The model also won't be able to download files from the internet for data analysis. What's interesting is that local functions, like user-uploaded images and conversation memory, will not be affected. This feature gives users more control over the balance between strong functionality and lower data leakage risk.