New ChatGPT features help you avoid data leaks
What's the story
OpenAI has introduced two new security features, Lockdown Mode and Elevated Risk labels, in its AI chatbot, ChatGPT. The company says these tools are meant to warn users about potentially risky features and limit external connections. This way, the chances of data leaks from prompt injection attacks can be minimized.
Security risk
What is prompt injection attack?
Prompt injection is a technique where hackers embed malicious instructions in web pages or files to manipulate an AI system into revealing confidential information or performing unintended actions. As millions use AI chatbots like ChatGPT for tasks such as document reading, web browsing, and app connections, the threat from these tricks becomes more pronounced. OpenAI's new tools aim to shield users from these risks.
Feature details
Lockdown mode restricts ChatGPT's external interactions
Lockdown Mode is an optional feature that tightly controls how ChatGPT interacts with external systems. When activated, it can restrict or disable certain tools and connections like live web browsing or integrations sending/receiving data from third-party services. The idea is to reduce these outside interactions and minimize the "attack surface" hackers could exploit.
User focus
Who should use Lockdown Mode?
OpenAI says Lockdown Mode isn't needed by most everyday users. It's mainly targeted at those who deal with highly sensitive information or think they might be at an elevated risk, such as journalists, executives, researchers, or security professionals. The company hopes this feature will give these high-risk users more control over their data and privacy while using ChatGPT.
Label introduction
Elevated risk labels for added transparency
Along with Lockdown Mode, OpenAI is also introducing Elevated Risk labels in ChatGPT. These labels will appear next to tools or features that expose more outside content or systems. For instance, if a feature connects to external content or provides broader system access, ChatGPT will show a clear warning about the potential risks. This way, users can decide whether to proceed with caution or keep things private.