OpenAI is replacing its screenshot-based Chronicle research preview with something potentially more invasive: Computer History, a feature that records your clicks, keystrokes, keyboard shortcuts, and app switches through macOS’s accessibility system, then converts them into a searchable ChatGPT memory timeline.
The pitch sounds useful. ChatGPT and Codex can pull from recorded activity to find recently edited documents, generate standup summaries, or suggest reusable automations when it spots repeatable workflows. Instead of screenshots or audio, the system captures interaction events and periodically converts them into text summaries stored as local Markdown files. That’s the tradeoff: no pixels, but a running transcript of what you did, when, and in which app.
Privacy controls exist, but they demand attention. Admins in Business and Enterprise workspaces must enable the feature before individual users can opt in, and Memories has to be switched on separately. Users can exclude specific apps, pause recording from the menu bar, and private browsing is supposedly never captured. But the memory files themselves sit unencrypted on the local file system. Any process running under the same macOS account could read them. OpenAI says the temporary event files get deleted after 48 hours and aren’t used for training—though when memories surface as context in later chats, those chats may still feed training data depending on account settings.
OpenAI’s own documentation flags prompt injection as a heightened risk, since malicious instructions embedded in a webpage could trick ChatGPT into following them. The company also advises against using Computer History with communication apps unless everyone involved consents. Europe is left out entirely for now—the feature isn’t available in the EEA, Switzerland, or the UK. That’s not a coincidence. This is the kind of always-on activity logging that European regulators tend to scrutinize hardest, and OpenAI appears to be testing the waters where scrutiny is thinner first.
