OpenAI Adds ChatGPT Lockdown Mode to Limit Prompt-Injection Data Exfiltration
Summary
OpenAI is rolling out an optional ChatGPT Lockdown Mode that limits web and external-service access to reduce prompt-injection data-exfiltration risks. It disables or restricts several features but does not eliminate all risk.
Key points
- Lockdown Mode is rolling out to eligible personal ChatGPT accounts and self-serve Business plans as an optional advanced security setting.
- It limits outbound network requests and external-service connections to reduce potential prompt-injection data-exfiltration paths.
- It restricts live browsing, image support, deep research, Agent mode, Canvas networking, and file downloads.
- Lockdown Mode does not prevent prompt injection or all its effects, and OpenAI says residual exfiltration risks may remain through enabled apps or new techniques.
- OpenAI also added session management tools to review active sessions and log out of individual or all sessions.
Article Details
- Event Type
- OpenAI rollout of an optional security mode and account session-management feature for ChatGPT
- Impact
- Lockdown Mode limits outbound network requests and disables or restricts features that could enable data exfiltration through prompt injection, including live browsing, image support, deep research, Agent mode, Canvas networking, and file downloads. It is rolling out to eligible personal and self-serve Business accounts. OpenAI says it reduces but does not eliminate exfiltration risk, and does not prevent all other effects of prompt injection. Users can also review active sessions and log out of individual or all sessions.