OpenAI is adding a stricter safety switch to ChatGPT
OpenAI has started rolling out Lockdown Mode, an optional security setting designed for people who want stronger protection against prompt injection attacks. The company says it is not meant for everyone, but for users and organizations handling sensitive data.
Prompt injection is one of the stranger threats in the AI era. Instead of attacking software in the usual sense, bad actors try to hide malicious instructions in webpages or other content so a chatbot can be tricked into following them. OpenAI says Lockdown Mode is meant as a last line of defense on top of the protections already built into ChatGPT, its models, and its backend systems.
For everyday users, the feature is less about adding new AI tricks and more about removing risky ones. That tradeoff matters because it shows where OpenAI sees the threat: not in what ChatGPT can say, but in what it can be coaxed into fetching or processing behind the scenes.
What Lockdown Mode changes
Turning on Lockdown Mode limits several ChatGPT features. Users can still generate images and upload photos, but the chatbot may not pull images from the internet or show images inside responses. It also cannot download files for analysis, although manual document uploads still work.
OpenAI says some tools are disabled completely when Lockdown Mode is active, including Deep Research and Agent Mode. Those cuts make the setting much more restrictive than a standard privacy toggle, which is exactly the point.
- Image generation still works
- Photo uploads still work
- Internet image retrieval may be blocked
- Images inside responses may be removed
- File downloads for analysis are disabled
- Manual document uploads still work
- Deep Research is disabled
- Agent Mode is disabled
What it does not change
OpenAI also says Lockdown Mode does not affect memory, file uploads, conversation sharing, or whether conversations may be used to improve models. The company notes that many of those options are separately configurable by workspace admins.
That detail is important for teams that use ChatGPT in controlled environments. Lockdown Mode is not a replacement for broader account or workspace policies. It is a tighter operating mode aimed at reducing data exfiltration risks connected to prompt injection, not a full reset of account behavior.
OpenAI further explains that the mode will not stop prompt injections from appearing in the content ChatGPT processes. Instead, the idea is to make it harder for an attacker to pull sensitive data out of an account by limiting the network requests that could be abused.
How to turn it on
Lockdown Mode is available to personal accounts, including users on OpenAI’s free tier. To enable it, open ChatGPT settings, go to Safety and security, then under Advanced security choose Lockdown mode and toggle it on.
If a user wants to drop back to normal settings for a specific conversation, OpenAI says they can do that temporarily by using the status message above the chat window and selecting Turn off for this chat.
OpenAI also adds a session manager
Alongside Lockdown Mode, OpenAI is rolling out an active session manager. It lets users see the devices and browsers currently associated with their account, then sign out of one session or all of them at once.
OpenAI says signing out of every session can take up to 30 minutes. If someone suspects unauthorized activity, the company recommends changing the password if one is used, reviewing sign-in methods, and contacting OpenAI Support.
Why this matters for ChatGPT users
This update shows how AI security is moving beyond simple account protection. As chatbots take on more web-connected tasks, the risk is no longer limited to stolen passwords or obvious scams. Prompt injection targets the model’s behavior itself, which makes the defensive response more specialized.
For most people, Lockdown Mode will probably stay off. But for companies, researchers, and anyone handling sensitive material, the option gives ChatGPT a more cautious profile. That may be the most useful part of the rollout: not broadening what the chatbot can do, but narrowing what it can be manipulated into doing.