OpenAI’s Lockdown Mode Cuts Off Web Access to Shield Sensitive Data from Prompt Attacks

OpenAI has rolled out a new security setting called Lockdown Mode, designed to help businesses and power users reduce the risk of prompt injection attacks. These attacks occur when malicious instructions are hidden inside webpages or other content sources, tricking a chatbot into leaking confidential information.

Lockdown Mode works by disabling several of ChatGPT’s most interactive features. Users can no longer browse live websites—only cached content is accessible—and the system will not retrieve or display images from the web, though image generation still works. Deep research and agent mode are also turned off.

Importantly, OpenAI acknowledges that Lockdown Mode is not a silver bullet. Prompt injections could still appear in cached web content or uploaded files, potentially affecting response accuracy. The goal, the company says, is to reduce the likelihood that sensitive data gets exfiltrated during a conversation.

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

The feature is currently rolling out to self-serve ChatGPT Business accounts and eligible personal accounts. For teams handling proprietary or regulated information, this offers a practical layer of defense without sacrificing core conversational capabilities.