OpenAI puts Lockdown Mode in ChatGPT as agent security gets practical

As AI agents touch the web, files, and external services, security controls are becoming part of the product surface. In its June 4, 2026 release notes, OpenAI made ChatGPT Lockdown Mode available to all logged-in users across account types and workspaces. The same release-note page also added moderation scores to generation requests in the Responses API and Chat Completions API.

Lockdown Mode is an optional advanced security setting aimed at reducing data-exfiltration risk from prompt injection. When it is enabled, ChatGPT restricts network-enabled capabilities such as live web browsing, deep research, agent mode, file downloads, and some web-derived image support. Personal users can turn it on under Settings > Security, while workspace admins can manage access through role-based controls.

The API change matters for developers building user-facing or agentic systems. OpenAI says developers can pass a moderation object in a generation request to receive moderation results for both model input and generated output in the same response. That can reduce the need for separate safety calls and make it easier to record risk signals at the exact point where the application decides what to show, store, or send to another tool.

The important shift is not a new benchmark score. It is a more operational question: which actions should a model be allowed to take, and which risk scores should be available before the application continues? As more products give models account, browser, or file access, buyers will ask for evidence that prompt injection and data leakage are controllable. Watch whether Lockdown Mode remains an optional switch or becomes a default requirement in regulated and high-risk workspaces.