- Computer History is off by default for Pro, Business and Enterprise users and is currently unavailable in the EEA, Switzerland and the U.K. [1] - It records interaction events rather than screenshots, screen recordings, microphone input or system audio. [1] - OpenAI says temporary event files are deleted after up to 48 hours, while generated Markdown memories remain locally until users clear them.
[1] OpenAI has introduced Computer History, a macOS feature that lets ChatGPT and Codex search a user’s recent activity across selected apps and websites. The system records events such as clicks, typing, keyboard shortcuts and application switches, then periodically converts them into text summaries and local memories. [1]
The feature is off by default for ChatGPT Pro, Business and Enterprise users. Pro users must opt in; Business and Enterprise customers also require administrator approval before individual users can enable it. Users can select included sources, exclude apps or websites, collection from the macOS menu bar, and delete individual entries or broader time ranges. [1]
Local files create the main security trade-off #
OpenAI says temporary event files are retained for up to 48 hours inside the ChatGPT app group and processed on its servers to create memories. The generated files are stored locally under the Codex memories directory as plain-text Markdown. OpenAI’s documentation warns that Computer History does not encrypt those files and that other programs running under the same macOS account may be able to access them. [1]
The feature does not capture screenshots, screen recordings, microphone input or system audio, and private-mode browsing is excluded. It can still collect activity from communication tools and websites, however, leading OpenAI to recommend pausing it during conversations unless participants have given consent and excluding apps containing health, financial or personal information. [1]
Training treatment depends on later use #
OpenAI says the temporary event files are not used for training. Its documentation adds that when a memory or related interaction event is later included as context in a ChatGPT or Codex conversation, that chat content may be used to improve OpenAI’s models if the user’s data-control settings allow it. OpenAI’s consumer policy says individual users’ prompts, responses and other submitted content may be used for model improvement depending on their settings. [1][3]
OpenAI also warns that Computer History increases prompt-injection risk because instructions embedded in websites or applications can enter the activity context available to ChatGPT or Codex. [1]
Companies mentioned #
Further sources #
The stories that matter, in one email. Free — unsubscribe anytime.