OpenAI Rolls Out Lockdown Mode to Shield Sensitive Info from Hidden Chatbot Tricks

OpenAI has introduced a new safety feature called Lockdown Mode, designed to guard against a growing threat known as prompt injection. This occurs when malicious instructions are hidden inside webpages or other content that a chatbot reads, potentially tricking it into revealing private data.

Lockdown Mode works by cutting off several of ChatGPT’s most interactive capabilities. Users won’t be able to browse the live web—only cached pages are accessible—and the system will stop fetching or displaying images from the internet (though it can still generate new ones). The feature also disables deep research and agent mode.

Still, OpenAI warns that Lockdown Mode isn’t a silver bullet. Even with it active, a prompt injection could slip through via cached content or an uploaded file, potentially altering how ChatGPT responds. The real goal is to reduce the odds that sensitive information gets leaked in the first place.

“Lockdown Mode is not intended for everyone,” the company states. “It is designed for people and organizations that handle sensitive data and want stricter protection from data exfiltration risks related to prompt injection.”

The rollout is currently underway for self-serve ChatGPT Business accounts and eligible personal accounts. For those handling confidential material, it’s a welcome layer of defense.