On August 13, 2026, OpenAI launched one of the most contentious decisions in its recent history: Computer History, an macOS activity tracker that logs clicks, keystrokes and application switching, transforming the user's daily flow into an AI-powered searchable timeline. The feature is embedded in the ChatGPT desktop app for Mac, requires version 2.6.0 or later, runs exclusively on Apple Silicon (M1 or newer) with macOS Sequoia 15.4 or higher and is available only to Pro, Business and Enterprise subscribers in supported regions — for corporate users, a workspace admin must enable access before any member can use it. The pitch sounds harmless: it is a digital memory that lets ChatGPT or Codex recover the context of something you were working on hours or days ago, without the need to re-explain every detail. The reality, however, is more complex than a simple task reminder. Computer History does not just log what you click; it monitors keyboard shortcuts, captures context that macOS's accessibility system exposes and converts these events into text summaries and local memory files. It does not capture screenshots, screen recordings, microphone input or system audio — but the amount of textual information flowing through macOS accessibility APIs is astonishingly vast, and it can include text from form fields, chat messages, document contents and even metadata from open files. OpenAI insists everything is stored locally as plain text, unencrypted, and that the feature is opt-in: you must explicitly enable it in the app's Settings menu under the "Integrations" section. In practice, however, the distinction between "opt-in" and "persuasive default" is razor-thin. The feature replaces "Chronicle," a Codex research preview that never reached maturity — suggesting OpenAI had been testing this premise for months. For users who work with dozens of applications open simultaneously, Computer History could be a productivity revolution. Imagine asking ChatGPT: "resume the report I was editing in Numbers and update the last table with the data I copied from Excel." The system has all the information about where you were, what you typed, which windows were visible. The problem is that the same information that makes the AI incredibly useful is exactly the same that makes it potentially invasive. When a third-party program — even one you chose to install — knows exactly which websites you visit, what text you type, what shortcuts you use and in what order, it is building a behavioral profile that goes beyond the product's scope. Computer History does not send data to the cloud in real time, which is a differentiator compared to Windows Recall, but memory files are generated periodically by the system itself and can be consulted at any time by the application, raising the question: for how long does this information stay local? If OpenAI decides in the future to use these files to train models or offer cross-device sync features, the boundary between "local storage" and "company data" will be thinner than it appears. There is also the security angle. Plain-text, unencrypted files stored on disk represent an attack surface that any malware with access to your user profile could exploit. If an attacker can read Computer History's memory files, they will have a chronological summary of your work sessions — in readable text, without encryption. The comparison with Windows Recall is inevitable. Microsoft's Recall, announced in 2024, captured continuous screenshots of the user's desktop and allowed AI to answer questions based on what you saw on screen. The result was a privacy massacre: users, journalists and security experts warned that the tool could access passwords, private messages, confidential documents and any content visible on the monitor. Recall was suspended for months after the backlash and relaunch in a scaled-down form a year later. OpenAI seems to have learned from the mistake, but only partially. By eliminating screenshots, making the feature opt-in and keeping everything local, the company avoided Recall's worst abuses. But what Computer History tracks — every click, every key pressed, every application you open — is, in terms of behavioral detail, just as invasive as a camera recording your screen. The difference is format, not intensity. The question is whether the convenience of having an AI that "knows what you were doing" is worth the price of delivering a complete digital mirror of your work behavior to a company that ultimately operates under a business model dependent on data to justify its pricing and its relevance. In a world where every digital interaction is potentially an asset, Computer History is not just a feature — it is a precedent.
Sources: The Verge, 9to5Mac, Windows Central
✓ Independent sources cross-checked and verified before publishing
