What if the AI you ask for help could watch everything you do first so it already knows the right answer?
That's the unsettling proposition behind ChatGPT's Computer History, a new opt-in feature in the desktop app for Mac users that logs clicks, keystrokes, and app switches to build a searchable timeline of a user's activity, according to The Verge. It's a feature that changes ChatGPT from a tool you consult into a persistent background process that observes your digital workday.
The core mechanism is simple: the app uses macOS's accessibility system to record your interaction events. It captures clicks, keystrokes, keyboard shortcuts, and app switches. OpenAI says it does not capture screenshots, screen recordings, microphone input, or system audio. Activity in private browser tabs is ignored.
These raw events are stored locally for up to 48 hours, then periodically sent to OpenAI's servers where a short-lived AI agent (Codex) summarizes them into a text "memory" file. This plaintext Markdown file is saved back to your Mac. The resulting timeline groups summaries by day and time, showing which apps were involved.
Why Is This More Than Just a Fancy Log?
On the surface, Computer History is a productivity booster. It lets you ask things like "what was I working on Tuesday?" or "show me the last document I edited" and get an accurate answer because ChatGPT has a memory of your actions, not just your chats. It can even spot repeatable workflows and suggest automations.
The deeper shift is from AI as a reactive tool to AI as an ambient observer. ChatGPT is no longer waiting for you to describe a problem; it is building a continuous model of your work from which it can infer problems and opportunities you haven't even articulated. As we’ve seen in OpenAI's push into premium business plans, the company is betting that deeper integration, not just smarter chatting, is the path to enterprise value.
What Does OpenAI's Own Documentation Warn You About?
The privacy trade-offs are significant, and OpenAI's own documentation highlights them.
First, the memory files this creates are stored on your Mac as unencrypted plaintext. This means any other program running under the same macOS user account could potentially read your activity summaries. On a personally managed Mac, that's a minor concern. On a company-owned or shared device, it's a data exposure vector that most current IT policies aren't equipped to handle.
Second, OpenAI flags an elevated prompt injection risk. Because Computer History captures content from apps and websites, malicious instructions embedded in a webpage or document could be recorded and later influence ChatGPT's behavior. They recommend excluding apps where you browse untrusted content.
OpenAI also warns against using the feature with communication apps unless everyone in the conversation has given explicit consent. For apps that handle sensitive health, financial, or personal data, OpenAI recommends pausing recording or excluding those apps entirely.
Third, activation is a multi-layered opt-in. For Business and Enterprise workspaces, an administrator must enable the feature first. Then, each individual user must also consent. The feature is currently unavailable in the European Economic Area, Switzerland, and the United Kingdom.
Who Should Actually Turn It On?
The utility versus risk calculation breaks down along clear lines.
- Turn it on if: You use ChatGPT for personal productivity on a personally managed Mac, and you can confidently exclude any apps holding sensitive data (financial, client, health info).
- Leave it off or be extremely selective if: Your work involves regulated or confidential data. The unencrypted local storage creates a real risk, especially on managed corporate devices. This is a scenario where the convenience of an AI assistant could introduce security vulnerabilities, similar to the issues that can arise when proprietary prompts are exposed.
For professionals, the critical step is reviewing the app inclusion list in Settings and removing any application where sensitive work occurs. If you're on a Business or Enterprise plan, check whether your admin has enabled it and understand your organization's stance before opting in.
Where Does This Lead? From Feature to Foundation
Computer History is not just another checkbox in settings. It's a foundational experiment in making the AI a layer of the personal computing environment itself. Its true test won't be whether people use it, but what behaviors it changes.
For the user, it creates a potential for radical efficiency. It also introduces a subtle pressure: the feeling of working under observation, which could shift open-ended exploration into logged, metrics-driven activity. Will an AI that knows your every digital move become the perfect assistant or a subtle governor of your workflow?
For OpenAI, it's a move toward creating immense switching costs. An AI that learns your deepest, most personal workflows becomes harder to replace than one that just answers your questions. It directly ties the company's growth to becoming an indispensable, intimate layer of your daily work, a strategy reflected in its aggressive revenue growth targets.
For the industry, Computer History is a shot across the bow. It shows what a tightly integrated AI-agent future looks like on the desktop. The question now is who follows, and how quickly. Will Apple, Google, and Microsoft respond with their own deeply integrated observation systems, or will they champion a different, perhaps more privacy-preserving, architectural approach like the on-device processing seen in some emerging models?
The final calculation is personal, and it's active. Enabling Computer History means blurring the line between using a tool and forming a symbiotic relationship with a digital entity that knows your work habits intimately. The value will be measured not just in minutes saved, but in the personal and professional cost of that intimacy. Watch closely to see if the initial quiet acceptance of this feature holds, or if a backlash builds as more users discover what their new digital shadow truly means.
Impact Analysis
- Introduces unprecedented surveillance capabilities directly into the daily tools consumers already trust, creating new privacy and security vulnerabilities at scale.
- Represents a fundamental shift from reactive AI assistants to proactive background observers, potentially altering workplace dynamics and autonomy without employee input.
- Raises critical questions about data ownership and use—activity logs are sent to OpenAI’s servers, blurring the line between local utility and corporate data harvesting.
Originally published on XOOMAR. For more news and analysis, visit XOOMAR.
Top comments (0)