OpenAI only announced Lock mode. Do you remember the lockdown? It’s like that, but with a chatbot with AI.
It is actually presented as a way to protect you from instant injection attacks.an insidious new type of AI-focused spam activity that is sometimes considered outright hacking. By embedding malicious instructions into content dumped in a message or found by an AI agent, attackers could have the ability to not only steal data, but also use it to try and take control of a user’s life.
Lockdown mode, then, is a simplified experience, as the name suggests. It’s not a block automatically triggered by danger, but rather a kind of panic room you enter when using ChatGPT if what you’re doing is so sensitive that a rapid injection attack could be especially damaging.
Essentially, every time the LLM breaks the containment of the small chatbot window in your browser or smartphone app and goes out to browse the Internet looking for information, retrieves an image, or, probably goes without saying, attempts to purchase airline tickets on your behalf, that opens you up to the danger of a quick injection attack, so lockdown mode disables those features.
Lockdown mode means that ChatGPT cannot:
- browse the web
- show images in responses (but it can generate images and you can upload images)
- do a “deep investigation”
- function as agent
- Network with the Canvas code generator.
- download files
As OpenAI says:
“The locking mode is No intended for everyone. “It is designed for individuals and organizations that handle sensitive data and want stronger protection against data breach risks related to rapid injection.”
AI chatbots are in their infancy and ChatGPT is creating a way aimed at preventing data theft. That’s reasonable, I guess. Still, it’s disconcerting to imagine, for example, lawyers downloading confidential information about their clients on ChatGPT, or doctors uploading their patients’ health data to their favorite LLM. There is an even more secure alternative “lockdown mode” to keep sensitive data safe from rapid injection attacks: Don’t let sensitive data anywhere near a chatbot.
