OpenAI is rolling out a new “Lockdown Mode” security setting for ChatGPT users, aimed at reducing the risk of data exfiltration tied to prompt injection attacks. Across reports, the feature is described as an optional control intended for people and organizations that handle sensitive information and want stronger safeguards over what ChatGPT can access and do when prompted by potentially malicious instructions.
Multiple outlets say Lockdown Mode limits the assistant’s ability to use external resources and certain product capabilities that could otherwise enable stolen data to be transmitted. This includes restrictions on live web browsing and other functions, though the exact list of disabled tools varies slightly by outlet. TechCrunch and Help Net Security note that the mode is designed to lower the likelihood of sensitive data being shared, while also indicating that vulnerabilities can still exist even with the added protections.
Availability is consistently reported for eligible logged-in personal accounts on Free, Go, Plus, and Pro, along with self-serve ChatGPT Business accounts in at least one report. ZDNet adds that enabling Lockdown Mode can reduce web access and therefore affects some user workflows. The rollout is described as being gradual, with access limited to users who qualify.
