OpenAI Unveils Lockdown Mode to Protect Sensitive Data from Prompt Injection Threats

OpenAI has launched a new feature to strengthen defenses against prompt injection attacks, which involve embedding harmful instructions into chatbots via webpages and other content sources.

Lockdown Mode, among other features, will disable live web browsing (allowing access only to cached content), prevent the retrieval and display of web images (although image generation will still be permitted), and restrict deep research and agent mode functions.

The company states that even with Lockdown Mode enabled, ChatGPT could still be vulnerable to prompt injections—these could arise, for example, through cached web material or uploaded files, potentially affecting the accuracy or conduct of a response.

Nonetheless, the purpose is to reduce the likelihood of sensitive information being accidentally disclosed.

“Lockdown Mode is not intended for everyone,” emphasizes OpenAI. “It is specifically designed for individuals and organizations that handle sensitive data and aim for greater protection against data exfiltration risks tied to prompt injections.”

The company is in the process of deploying Lockdown Mode to self-service ChatGPT Business accounts, as well as qualifying personal accounts.