OpenAI replaces screenshot-based Chronicle with keylogging-style Computer History for ChatGPT macOS users

OpenAI has shifted its approach to building ChatGPT memory timelines from Chronicle, which relied on screenshots, to Computer History, which captures input events such as clicks, typing, keyboard shortcuts, and app switches. The feature is opt-in and off by default, available to ChatGPT Pro, Business, and Enterprise users through the macOS desktop app. Pro users can enable it individually, while Business and Enterprise users require admin approval. It is explicitly not available in the EEA, Switzerland, or the UK, nor via API key or Amazon Bedrock.

The local storage model is a notable design choice with security implications. Computer History files are stored unencrypted, and OpenAI's own documentation acknowledges that other programs running as the same macOS user may be able to access them. Events are retained locally for up to 48 hours before deletion by ChatGPT and Codex, but they are also sent to OpenAI servers to generate memory summaries, which may be stored locally for longer periods and used in future chats that are passed back to OpenAI. OpenAI states it does not retain event files after processing unless required by law and does not use them for training, but it has provided chat logs in response to legal process in the past.

The feature introduces two concrete costs for users: increased token consumption from activity summarization and memory creation, and an expanded prompt injection attack surface. OpenAI explicitly warns that visiting a website containing malicious instructions could cause ChatGPT or Codex to follow those instructions. This is a direct acknowledgment that piping untrusted app and web content into the model's context window creates an execution path for adversarial prompts.

OpenAI's guidance to users reveals awareness of legal and ethical exposure. The company advises turning off Computer History during communications with other people unless prior express consent has been obtained, and suggests pausing it or excluding apps containing sensitive health, financial, or personal information. The feature does not capture screen images, microphone input, system audio, or private-mode browsing, which distinguishes it from the earlier Chronicle approach and from broader surveillance-style capture systems.

Only one feed carried this story, so there is no independent corroboration of the details beyond The Register's reporting on OpenAI's documentation. The framing is critical, emphasizing the surveillance implications of outsourcing activity capture to end users as a personal choice. For engineers evaluating whether to enable this in a Business or Enterprise deployment, the key tradeoffs are the unencrypted local storage, the server round-trip for event processing, the token cost overhead, and the prompt injection risk introduced by feeding untrusted content into model context.