ChatGPT’s Computer History tracks your clicks and keystrokes

By Adele Rutherford · August 17, 2026

There are certain mechanisms—certain procedural shortcuts, really—that should never have been invented.

When the Algorithm Becomes Your Witness

There are certain mechanisms—certain procedural shortcuts, really—that should never have been invented. They promise convenience, a seamless integration into the fabric of daily life, but in doing so, they strip away the very concept of private thought. OpenAI’s new "Computer History" feature for its ChatGPT desktop app is one such mechanism. It does not merely remember what you asked; it records how you lived your working day—every click, every keystroke, every application switch on your macOS machine. This isn't a helpful memory aid; it is the construction of an exhaustive digital ledger, and I submit that its very existence represents a profound failure of professional restraint by the tech sector.

The Unencrypted Ledger of Your Day

The facts are startlingly granular. As reported by elephas.app and theaicareerlab.com, Computer History functions by tapping into macOS’s accessibility APIs to track your raw activity—clicks, keystrokes, and app switches. It then runs an AI agent that processes this stream of events and converts it into a searchable "memory" file. The process is designed to help ChatGPT and Codex build skills or automations based on detected patterns in user activity. But the details are where the alarm bells should be ringing loudest: these memory files, which track your professional life across apps and websites, sit locally on your Mac as unencrypted plaintext.

The reporting from the-decoder.com makes this point crystal clear: any program running under the same macOS user account could potentially read those files. The developers have built a powerful tool that requires you to consent—it is opt-in, after all—but they have simultaneously created an infrastructure of vulnerability. They are asking you to hand over the keys to your digital life, not in one grand surrender, but through hundreds of tiny, seemingly innocuous interactions.

From Convenience to Predictive Profiling

The comparison drawn between this feature and Windows Recall is hardly accidental; both point toward a single, frightening trajectory: total behavioral capture. The mechanism—the systematic collection of every interaction event to build predictive profiles of association and movement—is not new. It echoes the architecture established during the era of Digital surveillance following 9/11. In that period, the state established comprehensive data collection infrastructure to build predictive profiles of citizens' movements and associations for preemptive security measures. The shared mechanism is unmistakable: the assumption that a sufficiently detailed record of your routine—your work habits, your sources of information, your collaborators—can be used not just to answer "what did you do," but to predict "what you will think."

The fact that OpenAI notes this feature does not capture images or audio, relying only on “events,” is merely a technical limitation; it does nothing to mitigate the fundamental threat. The data they are collecting is behavioral metadata—the most intimate and least defensible kind of information. It tells a story of your professional life without your ability to redact its inherent biases or mistakes.

The veneer of "user control"—the ability to exclude certain apps, for instance—is nothing more than window dressing on a fundamentally invasive system. The procedure itself is the danger: building an unencrypted, comprehensive map of one's digital existence that can be read by any program running under your account. This isn't about better productivity; it’s about creating a permanent, searchable record that bypasses traditional legal safeguards and professional boundaries.

This technology does not belong in the hands of consumer-facing AI products. The moment you allow an algorithm to become the witness to every keystroke, you have surrendered something essential: the right to operate without being perpetually observed. This system is too powerful, its architecture too invasive, and its security protocols too lax for anything but the most heavily regulated government use—and even then, only with warrants that specify a narrow scope of inquiry.

ChatGPT’s Computer History tracks your clicks and keystrokes

When the Algorithm Becomes Your Witness

The Unencrypted Ledger of Your Day

From Convenience to Predictive Profiling

Sources