OpenAI is locking down parts of ChatGPT to reduce data theft risks

OpenAI announced Lockdown Mode for ChatGPT, an optional security setting that applies sandboxing, blocks certain outbound network requests, and restricts features (web browsing, live connectors, Deep Research, Agent Mode, downloads) to reduce data exfiltration risks from prompt injection. The document explains the protections, lists affected capabilities and administrative controls for app and connector access, and clarifies limitations—Lockdown Mode limits network-based exfiltration but does not prevent prompt injection content or affect memory and file uploads.