ChatGPT adds PIN-protected chats with permanent deletion risk

Privacy for shared devices

OpenAI has begun rolling out a new feature that allows users to tuck sensitive conversations out of sight. These locked chats are hidden from the regular chat history, pinned conversations, and the general search function within the app. The primary goal is to protect private information when a user shares their screen or uses a device with other people.

To access these hidden conversations, users must enter a four-digit account PIN. On supported mobile devices, the app also supports unlocking these chats using Face ID, Touch ID, or the phone's standard device unlock method. These chats are organized in a specific area located under Projects and then Locked chats.

The company is gradually releasing this tool to a wide range of users. It is available for those on the Free, Go, Plus, Pro, Business, Enterprise, and Edu plans. While the feature works on the web and mobile apps, it is currently unavailable for voice interactions on the desktop app.

The risk of permanent deletion

The new security measure comes with a strict penalty for users who lose their credentials. If a user forgets their PIN and decides to reset it, OpenAI will permanently delete all content within the locked chats. This includes any files, content, or memories that were protected only by that specific PIN.

OpenAI stated that this deletion cannot be undone and the support team is unable to recover any lost content. Even workspace administrators do not have the power to reset a user's PIN or unlock their specific chats. This makes it essential for users to choose a PIN they can reliably remember.

In contrast, changing a PIN that a user still remembers does not result in any data loss. This distinction is important for users who want to update their security settings without risking their history. The system is designed to ensure that only a total loss of access triggers the permanent wipe of the data.

What locking does not do

It is important to understand that locking a chat changes where a conversation appears, but it does not change how OpenAI handles the data. The feature does not create a separate encrypted vault or provide end-to-end encryption. It also does not change the duration for which OpenAI retains the chat data or alter the user's settings regarding model improvement.

Furthermore, locking a chat does not remove any information that the AI had already remembered before the conversation was secured. It also cannot override existing workspace policies, safety requirements, or legal obligations. For instance, in Enterprise and Edu workspaces, locked chats are still included in the logs available through the Compliance Platform.

There are also specific limitations on which types of conversations can be locked. Users cannot lock temporary chats, archived chats, group chats, or chats created from shared links. Conversations with scheduled tasks, meeting recordings, and local Work conversations are also excluded from this new privacy feature.

Memory and data handling

Locked chats have specific rules regarding how the AI remembers information. When the Memory feature is active, information from new messages sent after a chat is locked can be remembered for use in other locked chats. However, this information will not carry over into the user's regular, unlocked conversations.

This distinction is significant because it separates the AI's 'memory' between different types of interactions. It allows for a degree of compartmentalization, ensuring that sensitive details shared in a locked environment stay within that specific context. This helps prevent the AI from mixing private data with everyday prompts.

Despite these visual and memory separations, the underlying data still follows the user's existing privacy settings. If a user has opted into the 'Improve the model for everyone' setting, their locked chats may still be used for training. Users who want to strictly limit how their data is used to train models must review their data controls separately.

The broader privacy landscape

The introduction of locked chats comes as AI chatbots become more integrated into daily life. OpenAI recently reported that ChatGPT now has 1.2 billion weekly users, highlighting the scale of data being processed. As the reach of these tools expands, questions about how companies handle personal information continue to grow for the general public.

Recent reports have shown that ChatGPT can search a Mac user's Apple Messages if granted permission. Other companies have faced similar scrutiny, such as when a trove of private conversations with Anthropic's Claude appeared in Google search results. These incidents have fueled a broader conversation about the safety of sharing personal information with AI models.

A study from the Stanford Institute for Human-Centered AI examined the privacy policies of six major U.S. developers, including Google, Meta, and Microsoft. The researchers found that all six appeared to use chat data to train and improve their models by default. The study raised concerns about long data-retention periods and a lack of transparency in how companies handle sensitive personal information.

Next steps for users

Users who want to utilize this feature should be cautious about the permanence of the PIN. Because the support team cannot recover data once a PIN is reset, it is advisable to write the PIN down in a secure physical location or a password manager. This ensures that the content remains accessible while staying hidden from the main interface.

For those concerned about how their data is used for training, the locked chat feature is not a substitute for privacy settings. Users should visit their account settings to review their data controls and model improvement preferences. This is the only way to ensure that their conversations are not used to train future versions of the AI.

As the technology evolves, users should keep an eye on how OpenAI updates its Compliance Platform and data retention policies. While locked chats offer a layer of visual privacy, the underlying data remains under the company's management. Staying informed about these policies is key to managing personal information in an AI-driven world.