Home / Technology / New AI Defense: Lockdown Mode Fights Prompt Injection
New AI Defense: Lockdown Mode Fights Prompt Injection
17 Feb
Summary
- Lockdown Mode enhances ChatGPT protection against prompt injection attacks.
- This feature limits AI interaction with external systems and data.
- Elevated Risk labels warn users of potentially risky AI tools and content.
OpenAI has launched Lockdown Mode to bolster security against prompt injection attacks, a significant risk for AI users, especially in professional settings. This feature enhances protection by restricting how ChatGPT interacts with external systems and sensitive data, thereby mitigating the risk of unauthorized access or data exfiltration.
Lockdown Mode is an optional setting primarily aimed at security-conscious users, including executives and security professionals within organizations using specific ChatGPT plans like Enterprise, Edu, Healthcare, and Teachers. The feature identifies and restricts high-risk tools and capabilities within ChatGPT to prevent exploitation of sensitive information.
In addition to Lockdown Mode, OpenAI is introducing Elevated Risk labels across its AI tools, including ChatGPT, the ChatGPT Atlas browser, and the Codex coding assistant. These labels serve as warnings for users when they access features with inherent risks, such as network access for Codex to search the web. While Lockdown Mode offers a robust defense, the Elevated Risk labels act as an immediate alert system, with plans for more comprehensive security enhancements in the future.
