OpenAI Unveils Lockdown Mode to Block AI Data Theft

Leading artificial intelligence company OpenAI has introduced a new advanced security feature aimed at preventing the theft of highly sensitive user information and data. The feature, called “Lockdown Mode,” is designed to protect ChatGPT from dangerous prompt injection cyberattacks. The move comes in response to the growing threat of hackers exploiting hidden or deceptive code embedded in malicious websites, emails, or uploaded documents to compromise AI chatbots.

The Trap of Prompt Injection and Data Theft

As AI adoption continues to accelerate in 2026, cybercriminals are becoming increasingly sophisticated. According to cybersecurity experts, prompt injection is a form of digital social engineering in the world of AI chatbots.

Hackers hide malicious instructions or prompts within webpages or documents that remain invisible to ordinary users. When ChatGPT scans such content during web browsing or data analysis, it may inadvertently prioritize the hidden instructions over the user’s original request. In the worst-case scenario, the chatbot could unknowingly transmit personal or organizational information from a user’s conversations to servers controlled by attackers.

How Lockdown Mode Works

OpenAI stated that Lockdown Mode primarily blocks outbound network requests—the final stage typically used by attackers to exfiltrate stolen information. When enabled, several major restrictions are imposed on ChatGPT’s functionality:

• Live Web Browsing Disabled: ChatGPT will no longer be able to retrieve new information directly from the live internet. Instead, it will be limited to viewing cached or previously stored content.
• Image Retrieval Suspended: The chatbot will not be able to fetch and display images from the internet in its responses. However, image generation through DALL·E and manual image uploads from a user’s device will remain available.
• Deep Research and Agent Mode Disabled: OpenAI’s advanced and widely used features, Deep Research and Agent Mode, will be completely unavailable while Lockdown Mode is active.
• Canvas and File Download Restrictions: Code created through ChatGPT’s Canvas feature will not be permitted to access the internet, and the chatbot will be prevented from downloading files directly for data analysis purposes.

Who Is This Feature For?

OpenAI clarified on its support page that Lockdown Mode is not intended for all users. Instead, it is designed primarily for corporate executives, legal and healthcare organizations, and security teams that regularly work with highly sensitive institutional information and are willing to sacrifice certain advanced capabilities in exchange for enhanced security.

The company also cautioned that Lockdown Mode does not provide absolute protection against prompt injection attacks. Hidden malicious instructions embedded in cached content or user-uploaded files may still affect the accuracy of ChatGPT’s responses. However, the feature significantly reduces the likelihood of sensitive data being transmitted to attackers, bringing the risk close to zero.

Broad Rollout Across Account Types

The feature is currently being rolled out to all personal ChatGPT account tiers, including Free, Plus, Pro, and self-service Business accounts. Users can enable Lockdown Mode by navigating to the Safety & Security section within ChatGPT settings.

OpenAI has also introduced the option to temporarily disable the feature for individual conversations when necessary. Alongside this update, the company has added an active session manager for Google account security, a measure intended to help prevent account compromise and unauthorized access.

DBTech/BMT/OR