OpenAI launches Lockdown Mode to enhance ChatGPT security
- Thảo Nguyên
- Jun 8
- 2 min read
Data security is becoming a top priority for tech users worldwide. To mitigate data leakage risks from cyberattacks, OpenAI has recently rolled out an advanced security feature called Lockdown Mode for eligible personal ChatGPT accounts.
What is ChatGPT's Lockdown Mode?
This is an optional security setting designed to restrict tools and features capable of connecting to the internet or external services within the OpenAI ecosystem. The core objective of this feature is to minimize the risk of sensitive data exfiltration arising from prompt injection attacks - a common attack vector targeting Large Language Models (LLMs).
This feature is specifically designed for individuals and organizations that regularly handle sensitive data requiring stringent protection measures. Logged-in users across various service tiers - including Free, Go, Plus, Pro, as well as the self-service ChatGPT Business plan - can access this tool.
How it works and restricted features
Technically, this solution is built upon a secure sandbox environment and existing URL data leak controls. The system restricts outbound network requests to block the transmission of critical data to attacker-controlled infrastructure.
It is important to note that the purpose of Lockdown Mode is not to completely prevent prompt injection attacks from occurring, nor does it interfere with memory, file upload capabilities, or conversation sharing permissions. Instead, it operates by severing the pathways through which data could be exfiltrated.
In exchange for this enhanced security, several useful features will be disabled or restricted, including:
Live web browsing: Internet access is restricted; the system only allows viewing cached content.
Image support: Disables rendering images in standard responses or fetching images from the web.
Deep research.
Agent mode.
Canvas network connectivity.
File downloads.
Important security considerations
OpenAI emphasizes that this security feature is not intended for the general user base. A notable limitation is that users cannot simultaneously enable Lockdown Mode and Developer Mode - activating one will automatically disable the other.
Furthermore, while this solution significantly mitigates data breach risks, it cannot guarantee 100% absolute security. Vulnerabilities may still emerge through active applications, unforeseen feature combinations, or novel attack techniques.
Lockdown Mode also cannot prevent all other impacts of prompt injection attacks. For instance, if an uploaded file contains embedded malicious instructions, it may still manipulate ChatGPT's behavior and cause the AI to generate misleading or inaccurate responses.
How to manage your account more securely?
If you suspect unauthorized access, you can proactively log out of individual sessions or terminate all sessions simultaneously. Detailed information listed by the system includes:
Device type and app in use.
Estimated geographic location.
Detailed login date and time.
Device trust verification status.
Current session marker.
The addition of Lockdown Mode alongside session monitoring features demonstrates OpenAI's proactive steps toward strengthening security barriers. While users must accept a reduction in certain convenient features, it is a necessary trade-off to safeguard information in today's digital landscape.
Reference: The Hacker News
Comments