OpenAI introduced a new security feature called Lockdown Mode for ChatGPT, designed to protect sensitive information by significantly restricting the AI's external connectivity. Available across all account tiers as of June 6, 2026, the feature is specifically intended for individuals and organizations that handle highly confidential data. The primary function of Lockdown Mode is to mitigate the risks associated with prompt injection, a security vulnerability where malicious instructions are embedded within documents, websites, or emails to manipulate an AI model's behavior. While OpenAI notes that the mode does not remove malicious code from uploaded content, it aims to prevent the exfiltration of sensitive information by disabling external network access.

When Lockdown Mode is active, ChatGPT transitions into a more isolated state. Live web browsing is effectively shut down, forcing the system to rely only on cached content and potentially limiting the availability of up-to-date search results. Furthermore, the feature disables advanced capabilities including Deep Research, Agent Mode, and network access for code generated through Canvas. The system also loses the ability to download external files for analysis. While users can still upload images or generate visuals, the AI cannot fetch external images from the web. These restrictions prioritize security over the convenience of a highly connected assistant, mirroring the approach taken by Apple's own security-focused Lockdown Mode.

The release reflects a broader industry shift toward scrutinizing the level of access AI systems should possess as they gain capabilities to read internal documents, browse the internet, and connect to business software. OpenAI is positioning this as a choice for users, acknowledging that total protection is difficult because prompt injection remains an ongoing research challenge. For enterprise customers, the feature offers additional granular control, allowing workspace administrators to create custom roles, restrict specific apps or connectors, and limit write-enabled integrations that could create pathways for data to leave secure environments. OpenAI suggests that as AI assistants become more powerful, tools that allow users to dial back those capabilities will become increasingly essential for maintaining organizational security.

OpenAI introduced a new security feature called Lockdown Mode for ChatGPT, designed to protect sensitive information by significantly restricting the AI's external connectivity. Available across all account tiers as of June 6, 2026, the feature is specifically intended for individuals and organizations that handle highly confidential data. The primary function of Lockdown Mode is to mitigate the risks associated with prompt injection, a security vulnerability where malicious instructions are embedded within documents, websites, or emails to manipulate an AI model's behavior. While OpenAI notes that the mode does not remove malicious code from uploaded content, it aims to prevent the exfiltration of sensitive information by disabling external network access.

When Lockdown Mode is active, ChatGPT transitions into a more isolated state. Live web browsing is effectively shut down, forcing the system to rely only on cached content and potentially limiting the availability of up-to-date search results. Furthermore, the feature disables advanced capabilities including Deep Research, Agent Mode, and network access for code generated through Canvas. The system also loses the ability to download external files for analysis. While users can still upload images or generate visuals, the AI cannot fetch external images from the web. These restrictions prioritize security over the convenience of a highly connected assistant, mirroring the approach taken by Apple's own security-focused Lockdown Mode.

The release reflects a broader industry shift toward scrutinizing the level of access AI systems should possess as they gain capabilities to read internal documents, browse the internet, and connect to business software. OpenAI is positioning this as a choice for users, acknowledging that total protection is difficult because prompt injection remains an ongoing research challenge. For enterprise customers, the feature offers additional granular control, allowing workspace administrators to create custom roles, restrict specific apps or connectors, and limit write-enabled integrations that could create pathways for data to leave secure environments. OpenAI suggests that as AI assistants become more powerful, tools that allow users to dial back those capabilities will become increasingly essential for maintaining organizational security.