OpenAI announced a brand new characteristic that it says will present further safety from immediate injection assaults, the place malicious chatbot directions are hidden in webpages and different content material sources.
Amongst different issues, Lockdown Mode will disable dwell internet shopping (so you may solely entry cached content material), the retrieval and show of pictures from the online (you may nonetheless generate pictures), deep analysis, and agent mode.
The corporate says that even with Lockdown Mode turned on, ChatGPT may nonetheless be weak to immediate injections — which may, for instance, “seem in cached internet content material or in an uploaded file, and will nonetheless have an effect on the habits or accuracy of a response.”
However the purpose is to cut back the probability that delicate information will get shared within the course of.
“Lockdown Mode is not supposed for everybody,” OpenAI says. “It’s designed for individuals and organizations that deal with delicate information and need stricter safety from information exfiltration dangers associated to immediate injection.”
The corporate says it’s presently rolling Lockdown Mode out to self-serve ChatGPT Enterprise accounts, in addition to eligible private accounts.
