OpenAI's latest ChatGPT feature records your keystrokes and saves them in unencrypted text format.

      OpenAI has introduced a ChatGPT feature called Computer History, which logs clicks, keystrokes, keyboard shortcuts, and app switches on macOS to create a searchable memory. The memory files are stored locally as unencrypted plain text, and this feature is not offered in the EEA, Switzerland, or the UK.

      Computer History provides ChatGPT with a record of activities on your Mac. It logs interactions through the accessibility framework of the operating system and transforms them into a memory that the assistant can access.

      However, its capabilities are more limited than they appear. The feature does not capture screenshots, screen recordings, microphone input, or system audio, and it does not apply to private browsing.

      This limitation is a purposeful adjustment. Computer History replaces the earlier Chronicle feature from Codex that used screen captures for context, instead opting for input events.

      One aspect remains unchanged: the memory files are stored in plain-text Markdown locally without encryption, which means any application operating under the same macOS account can read them.

      The consent process is more nuanced compared to most launches. In Business and Enterprise workspaces, an administrator must enable the feature, the individual user must give separate consent, and Memories must be activated.

      Users have tangible controls as well. They can create lists to include or exclude specific apps and websites, and they can pause recording from the menu bar.

      On the raw end, data retention is brief. Temporary event files remain on the device for 48 hours before being deleted, and OpenAI states that they process these files on their servers without storing them afterwards, unless legally obligated.

      The training aspect presents some ambiguity. OpenAI claims that memory files are not utilized for model training; however, if memories are referenced in future chats, the contents of those chats may still be used for training data based on user settings.

      The company's documentation is notably candid about potential risks. It highlights an increased risk of prompt injection, cautions against using the feature with communication apps without agreement from all parties involved—reflecting an ongoing industry issue regarding recording others—and recommends excluding any information related to health, financial, or personal matters.

      Users in Europe are unable to access this feature, which is not available in the European Economic Area, Switzerland, and the United Kingdom.