Executive Summary
OpenAI has launched two new security features, Lockdown Mode and "Elevated Risk" labels, to protect users from prompt injection attacks. These tools address the increasing risks associated with AI interacting with external apps and the web. Lockdown Mode provides a highly restrictive setting for security-conscious enterprise users by limiting features like live web browsing, while Elevated Risk labels clearly identify potentially risky capabilities across OpenAI's products to help users make informed decisions.
Key Takeaways
* Lockdown Mode: An optional, advanced security setting for enterprise users that tightly constrains ChatGPT's interaction with external systems to prevent data exfiltration.
* How Lockdown Mode Works: It limits web browsing to cached content (no live network requests) and disables certain tools entirely to provide deterministic guarantees of data safety.
* Admin Controls: Workspace Admins can enable Lockdown Mode for specific user roles and can still approve a granular list of apps and specific actions within those apps.
* "Elevated Risk" Labels: A standardized visual indicator applied to capabilities across ChatGPT, ChatGPT Atlas, and Codex that may introduce additional security risks (e.g., granting network access). These labels provide clear in-product guidance.
* Availability: Lockdown Mode is now available for ChatGPT Enterprise, Edu, Healthcare, and Teachers plans. It is planned for consumer release in the coming months.
Strategic Importance
This announcement signals OpenAI's focus on enterprise-grade security and trust as its models become more integrated with external data and applications. By proactively addressing vulnerabilities like prompt injection, OpenAI is positioning its platform as a more secure and reliable choice for handling sensitive workflows.