Tech Souls, Connected.

OpenAI adds Lockdown Mode for users handling sensitive information

OpenAI has unveiled Lockdown Mode, a new security feature designed to provide stronger protection against prompt injection attacks.

The company said the feature is aimed at users and organizations that handle sensitive information and want to reduce the risk of data exposure.

Prompt injection attacks involve malicious instructions hidden in web pages or other content sources that can influence how a chatbot responds.

Features restricted in Lockdown Mode

When enabled, Lockdown Mode disables several capabilities, including:

  • Live web browsing, allowing access only to cached content.
  • Retrieval and display of images from the web.
  • Deep research.
  • Agent mode.

Users can still generate images while the setting is active.

Protection is not absolute

OpenAI said Lockdown Mode does not eliminate the threat of prompt injection attacks.

According to the company, malicious instructions could still be present in:

  • Cached web content.
  • Uploaded files.

Such content could still affect the behavior or accuracy of ChatGPT’s responses.

The purpose of the feature is to lower the chances that sensitive information is exposed during interactions.

Designed for high-security use cases

“Lockdown Mode is not intended for everyone,” OpenAI said.

“It is designed for people and organizations that handle sensitive data and want stricter protection from data exfiltration risks related to prompt injection.”

Rollout underway

OpenAI said it is currently rolling out Lockdown Mode to:

  • Self-serve ChatGPT Business accounts.
  • Eligible personal accounts.

TL;DR:

OpenAI has launched Lockdown Mode, a security feature that disables live web browsing and several advanced capabilities to reduce the risk of prompt injection attacks and protect sensitive information.

AI summary:

  • OpenAI introduced Lockdown Mode for ChatGPT.
  • The feature aims to reduce prompt injection risks.
  • Live web browsing, deep research and agent mode are disabled.
  • Prompt injection attacks are still possible through cached content or uploaded files.
  • Lockdown Mode is being rolled out to ChatGPT Business and some personal users.
Share this article
Shareable URL
Prev Post

Former Andreessen Horowitz partner exits Trump administration

Next Post

From Social Security fears to ransomware: 2026’s worst hacks

Read next