OpenAI has released ChatGPT Computer History, a new feature for macOS that logs a user’s clicks, keystrokes, keyboard shortcuts and app switches to build a searchable memory. The recordings are stored locally as unencrypted plain text, and the feature is not available in the European Economic Area, Switzerland or the United Kingdom.
The design echoes, and diverges from, Microsoft‘s Recall feature for Windows. Recall drew criticism in 2024 for storing periodic screenshots of user activity, and Microsoft delayed its rollout to add opt-in controls. OpenAI’s local, unencrypted storage model raises a similar question: any process with access to the same Mac account can still read the plain-text logs, even without cloud exposure.
Computer History runs through the operating system’s accessibility framework, turning raw input events into a memory that ChatGPT can draw on later. It replaces Chronicle, an earlier feature in Codex that captured screenshots to build context. The new version trades images for input events, a narrower approach than its predecessor.
What ChatGPT Computer History does and doesn’t capture
The feature takes no screenshots, no screen recordings, no microphone input and no system audio. Private browsing sessions are excluded from the log. Users can add specific apps and websites to include or exclude lists, and recording can be paused directly from the menu bar.
The gap sits in storage. OpenAI keeps the memory files as plain-text Markdown on the local machine without encryption, so any other program running under the same macOS account can read them.
Consent rules for business accounts
On Business and Enterprise workspaces, an administrator has to turn the feature on before anyone can use it. After that, each individual user has to opt in separately, and the broader Memories setting also has to be switched on. Temporary event files stay on the device for 48 hours before deletion, and OpenAI says it processes them on its servers without retaining them afterward, unless required by law.
Training data is where the policy gets less clear cut. OpenAI says the memory files themselves are not used to train its models. However, when a memory surfaces as context inside a later chat, that chat’s contents may still be used for training, depending on the user’s account settings.
OpenAI flags its own risks
OpenAI’s documentation for Computer History is direct about the downsides. The company warns of heightened exposure to prompt injection attacks, and it advises against using the feature inside communication apps unless everyone in the conversation has agreed to be recorded. It also recommends excluding any app that handles health, financial or personal data.
The feature remains unavailable across Europe. Users in the EEA, Switzerland and the UK cannot enable Computer History under current regional rules.
- No screenshots, screen recordings, microphone or system audio capture
- Private browsing excluded by default
- Temporary files deleted from the device after 48 hours
- Unavailable in the EEA, Switzerland and the UK