ChatGPT Locked Chats feature

When OpenAI rolled out Locked Chats on October 20, 2026, it filled a real gap in ChatGPT's privacy toolkit - a way to hide sensitive conversations from anyone glancing at a shared screen or borrowed device^[1]. But independent commentary converged almost immediately on the same caveat: the feature controls visibility, not data handling. OpenAI's own support documentation states plainly that locking a chat does not create end-to-end encryption or a separate encrypted vault, does not change retention, and does not change the 'Improve the model for everyone' setting^[2]. aiweekly.co bluntly labeled the design 'a shoulder-surfing UX, not a privacy upgrade'^[3], while Absolute Geeks drew the same line between device-level protection and server-side security, calling it 'a privacy screen for your device... not a guarantee that your data is shielded on OpenAI's side'^[4]. ExplainX pushed the analogy further, writing that 'Locked Chats is a door lock inside your house, not a safe... It does not stop the landlord, who is the provider, from having a key under its own policy'^[5].

The more concrete risk sits one step removed from the encryption debate: what happens when a user simply forgets their PIN. Resetting a forgotten account PIN permanently deletes all locked chats, any memories stored only inside those locked chats, and any files protected only by the old PIN, with no recovery path^[6]. Changing a PIN a user still remembers leaves locked conversations untouched; it is only the forgot-and-reset flow that triggers permanent loss^[6]. That design choice makes sense from a security standpoint, since a recoverable PIN would be a weaker PIN, but it is a sharper failure mode than most consumer privacy features carry, given that Locked Chats - unlike the ephemeral Temporary Chat mode it is often compared to - is meant to be a persistent, returnable space^[1]. Temporary Chat itself is not a clean privacy baseline either: OpenAI retains a copy of those sessions for up to 30 days for safety purposes before deletion^[8], a reminder that no mode in ChatGPT today offers a true walk-away from the provider's own copy.

Because the feature shipped only a day before this research was compiled, its real-world reception is still forming. Online chatter has been muted and mostly procedural rather than analytical - the loudest recurring thread was not about privacy mechanics at all, it was people comparing notes on whether they had actually received the feature yet, with a visible split between users who saw Locked Chats appear and others still waiting on it to arrive. That pattern lines up with OpenAI's own description of a staged, cross-platform rollout spanning web, iOS, Android, and new desktop apps for macOS and Windows rather than a simultaneous flip for every account^[1]. On X, the shape of the reaction was similar in tone: OpenAI's own @ChatGPT account announced the feature to broad, mostly positive-to-neutral pickup, with accounts like @testingcatalog reporting on which subscription tiers were receiving the rollout and @RoundtableSpace describing how the lock-and-unlock mechanics worked for their followers. The sharpest pushback on the platform - questions about why the feature stops at a PIN rather than end-to-end encryption - tended to surface in replies underneath that coverage rather than in the original posts themselves, mirroring the same visibility-versus-security distinction independent outlets were drawing in their own writeups.

Locked Chats is less a new idea than an established consumer-privacy pattern arriving late to ChatGPT. WhatsApp has offered a comparable Chat Lock feature with a secret code for protecting individual conversations from casual device access^[7], and OpenAI's version - PIN or biometric unlock, chats removed from the normal list view - mirrors that same shared-device threat model almost feature for feature. The move reads less as OpenAI inventing a new privacy primitive and more as it extending a now-familiar messaging-app UX pattern into a product that millions increasingly use to discuss deeply personal material.

One genuinely new piece of plumbing is the memory partition. When ChatGPT's Memory feature is on, Locked Chats gets its own separate memory pool shared only among conversations inside that section; anything a user tells ChatGPT after locking a chat updates that isolated memory, not their regular Memory, so information discussed in a locked chat cannot quietly surface in an unlocked one^[1]. That is a real architectural change, not just a UI hide-and-show trick, though OpenAI is careful to note it does not retroactively scrub anything ChatGPT had already remembered before the chat was locked^[2].