OpenAI adds Lockdown Mode to protect ChatGPT data

OpenAI has begun rolling out a new security feature called Lockdown Mode for its ChatGPT service, designed to reduce the risk of sensitive data being stolen through prompt injection attacks. The optional setting is now available to eligible personal accounts, including free and paid tiers, as well as self-serve ChatGPT Business and managed enterprise workspaces.

Prompt injection attacks involve hiding malicious instructions within webpages, uploaded files, or other content that an AI model processes. These attacks can manipulate the chatbot's behavior or lead to inaccurate responses. While a significant security challenge across the industry, Lockdown Mode specifically aims to block the final stage of such an attack: the unauthorized transfer of data out of a user's account.

The company stated that Lockdown Mode is not intended for all users but is designed for individuals and organizations handling sensitive data who require stricter protection. When enabled, the feature limits or disables several capabilities that connect to external services to restrict outbound network requests. These disabled functions include live web browsing, the Deep Research feature, and Agent Mode....