What happened
OpenAI introduced "Lockdown Mode" for ChatGPT, an optional security setting designed to protect sensitive data from prompt injection attacks. This mode disables live web browsing, image retrieval and display from the web, deep research, and agent mode, limiting ChatGPT's interaction with external systems. OpenAI states that while it reduces data exfiltration risks, it does not eliminate all prompt injection vulnerabilities, as attacks can still occur via cached content or uploaded files. The feature is rolling out to self-serve ChatGPT Business accounts and eligible personal accounts.
Why it matters
Lockdown Mode provides a new control for organisations handling sensitive data, reducing specific data exfiltration vectors. Security architects and procurement teams gain a mechanism to mitigate prompt injection risks by limiting outbound network requests and disabling certain connected features. This offers a configurable defence, though OpenAI notes that prompt injection vulnerabilities can persist in cached content or uploaded files, requiring continued vigilance. This follows persistent reports of prompt injection attacks remaining unfixed, highlighting the ongoing challenge for AI security.
