ChatGPT Lockdown Mode is an optional advanced security setting that limits some outbound network requests to help reduce the risk of sensitive data being exfiltrated through prompt injection. The trade-off is that features such as live web browsing and Deep Research are restricted. It does not guarantee that data cannot be exfiltrated or prevent malicious instructions from affecting an answer.
Eligible personal and self-serve ChatGPT Business users can enable it when the setting is available; administrators control it through roles in managed workspaces. This guide is based on OpenAI's help documentation, checked on November 7, 2026. Availability depends on the plan, workspace, product surface, rollout, and administrator settings. The steps below describe the official path; we have not performed them in an account.
What Lockdown Mode does and limits
Prompt injection is an attempt to change an AI system's behavior using instructions hidden in content such as a web page or file. Lockdown Mode primarily limits outbound requests that could send data to an attacker. Malicious instructions can still appear in cached pages and uploaded files and affect ChatGPT's behavior or answer accuracy. Enabled apps, unforeseen combinations of capabilities, and new techniques may still create exfiltration risks. Follow your organization's data and access policies when handling sensitive information. See OpenAI's risk explanation.
OpenAI describes these main feature changes:
| Feature | Change |
|---|---|
| Web browsing | Limited to cached content; search results may be limited, unavailable, or stale |
| Images | Images may not appear in regular responses or be retrieved from the web; image uploads and otherwise available image generation remain available |
| Deep Research, Agent Mode | Disabled |
| Canvas networking | Users cannot approve Canvas-generated code to access the network |
| File analysis | ChatGPT cannot download files for data analysis; it can still work with manually uploaded files |
Lockdown Mode does not change memory, file uploads, conversation sharing, or whether conversations may be used to improve models; these may need separate configuration. It also does not affect network access in Codex. Consider the trade-off before enabling it if your work depends on live web access or disabled features. Source: OpenAI Help Center
Turn it on for a personal or self-serve Business account
Sign in to an eligible account first. Where the option is available, the following path applies to personal and self-serve ChatGPT Business accounts:
- Open Settings and select Security.
- Under Advanced security, turn on Lockdown Mode.
- Select Turn on in the confirmation dialog.
If you cannot find the option, first check that you are signed in to the intended account and workspace. The feature also depends on your plan, product surface, rollout, and administrator settings, so the missing switch alone does not identify the cause. Organization members can ask their administrator about workspace policy. Source: availability and setup
Turn it off or back on for one chat
For accounts that can set it themselves, select the LOCKDOWN MODE label in the web composer and choose Turn off for this chat. Alternatively, open the chat's ••• menu and select Lockdown > Disabled; to restore it, select Lockdown > Enabled in the same menu. These choices change only the current chat. Manage lockdown in the composer dialog opens the account's Security settings. Source: OpenAI Help Center
Turning it off temporarily removes the Lockdown restrictions for that conversation. Before continuing with sensitive information, check that the tools and data you need meet your security requirements.
Configure and troubleshoot a managed workspace
In a supported managed workspace, an administrator can create a custom role, designate it as a Lockdown Mode role, and assign members or groups to it. Members who need a restricted capability should contact their workspace administrator; do not assume the personal-account chat control overrides an administrator-assigned role. Source: OpenAI Help Center
Apps and MCPs in managed workspaces are governed by workspace settings and role-based access controls. Lockdown Mode does not automatically disable every app. Administrators should enable only trusted apps and actions needed by members using Lockdown Mode, and assess their exfiltration risk. OpenAI does not recommend read or write actions for untrusted apps, or write actions for trusted apps with broad or uncertain visibility. Read actions from trusted apps can still expose sensitive data. Source: app risk guidance
If a member cannot use an app, MCP, or action, an administrator can check:
- Review roles assigned directly and through groups. Ordinary role permissions generally combine additively, but a Lockdown Mode role can still restrict a capability granted by another role.
- Check whether the app is assigned to the member, group, or role and whether the necessary read or write action is enabled.
- Check whether the connected account can access the relevant file, repository, channel, or record in the source system. ChatGPT app access does not override source-system permissions.
- If a Lockdown Mode role restriction needs to be removed, the workspace owner should handle the member's or group's role assignment according to organizational policy.
These checks follow OpenAI's roles and permissions guidance, app management and troubleshooting guidance, and Lockdown Mode documentation. App behavior differs for personal and self-serve Business accounts: Lockdown Mode blocks live access to external services or app write actions when they conflict with its protections, and some connected experiences are unavailable.