June 5, 2026
Leak panic hits the chatbot
OpenAI Help: Lockdown Mode
OpenAI just added a panic button, and users are asking what it says about the default setup
TLDR: OpenAI added Lockdown Mode, a stricter setting that turns off several web-connected ChatGPT features to reduce the chance of sensitive data being tricked out of the system. Commenters are split between praising the extra protection and asking the uncomfortable question: why wasn’t the normal setup strong enough already?
OpenAI has rolled out Lockdown Mode, a new optional safety setting for ChatGPT that basically says: if you’re handling sensitive stuff, maybe let the bot do less. Turn it on, and ChatGPT loses a bunch of internet-connected powers like live web browsing, deep research, agent mode, and file downloading. The idea is simple in plain English: if a sneaky prompt hidden in a webpage or file tries to trick the AI into leaking your data, there are fewer ways for that data to escape.
But the comment section immediately turned this from a dry help doc into a full-on trust drama. One camp called it the obvious, smart move and said companies will love it. Another camp went straight for the awkward implication: if Lockdown Mode is needed, does that mean normal mode isn’t safe enough? That was the real gasp-inducing takeaway. One commenter bluntly said the feature’s very existence suggests the default protections may not stop a determined attacker.
Then came the eye-rolls and memes. People noticed the name sounds a lot like Apple’s Lockdown Mode and joked that OpenAI may be borrowing the vibe. Others zoomed in on the fine print and found a juicy catch: Codex isn’t covered, meaning one important path to the outside world still needs separate controls. And the snarkiest hot take of all? A commenter basically said, “Cool, we still haven’t solved the core problem, we’ve just built a smaller box around it.” In other words: applause, suspicion, and a whole lot of “wait… what exactly was happening before?”
Key Points
- •OpenAI introduced Lockdown Mode as an optional security setting to reduce data exfiltration risk from prompt injection attacks by limiting outbound network requests.
- •The feature is rolling out to eligible personal accounts, including Free, Go, Plus, and Pro, as well as self-serve ChatGPT Business accounts.
- •Lockdown Mode disables or restricts several capabilities, including live web browsing, deep research, agent mode, Canvas network access approvals, and file downloads for data analysis.
- •OpenAI states that Lockdown Mode helps prevent the final exfiltration stage of prompt injection attacks but does not stop prompt injections from appearing in cached content or uploaded files.
- •Lockdown Mode does not change memory, file uploads, conversation sharing, or model-improvement settings, and it does not affect network access in Codex.