OpenAI began rolling out a feature called locked chats this week, letting ChatGPT users secure individual conversations behind a four-digit PIN. Locked conversations disappear from regular chat history, pinned chats, and search results, and move instead into a dedicated section under Projects, Locked chats, where they stay out of view until unlocked. On mobile, users get a faster option: Face ID or Touch ID instead of typing a PIN.
What it actually protects against
The feature is built for a specific kind of exposure: someone else picking up your phone or laptop and scrolling through your chat history. That's a real and common privacy gap, and locking a chat genuinely closes it. What it doesn't do is change anything about where that conversation's content goes afterward. Locked chats aren't end-to-end encrypted, OpenAI's data retention policies are unchanged, and the company's use of conversations to train and improve its models continues exactly as before. The lock is a visibility control, not a data control.
The one feature detail worth reading twice
There's a sharp edge built into the design: if you forget your PIN and have to reset it, every locked chat tied to that PIN is permanently deleted, content included, and OpenAI says the loss cannot be undone, with its own support team unable to recover it. That's an unusually hard tradeoff for a feature marketed as a safety net, forgetting a PIN doesn't just lock you out, it erases the thing you were trying to protect.
Where it's available, and where it isn't
The rollout covers ChatGPT's full plan lineup, Free, Go, Plus, Pro, Business, Enterprise, and Edu, and spans the new macOS and Windows desktop apps, the web version, and both iOS and Android apps. The feature has real gaps though: temporary chats, archived conversations, group chats, and several other chat types can't be locked at all, which means the conversations some users most want hidden may not be eligible for the protection in the first place.
The backdrop this is landing against
The timing isn't incidental. A Stanford study cited alongside the rollout found that major AI developers, OpenAI included, use users' chat data to train and improve their models by default, the exact practice locked chats leaves untouched. OpenAI has also already been through one public privacy stumble with ChatGPT specifically, after shared conversations started turning up indexed in Google search results, a problem the company fixed by removing that indexing option rather than by rethinking sharing controls from the ground up.
Why this matters beyond the feature itself
For anyone building marketing or customer-facing products on top of large language models, locked chats is a useful case study in the gap between a privacy feature and a privacy promise. A PIN that hides content from a screen is real, visible, demonstrable value to a user worried about a shoulder surfer. A PIN that doesn't touch retention, training use, or encryption is a much narrower claim than most users will assume from the word locked, and the companies shipping these features are increasingly betting that the UI metaphor does more persuading than the fine print does explaining.
