OpenAI has released a new ChatGPT feature for Mac called Computer History. It quietly logs your clicks, your typing, your keyboard shortcuts, and which apps you jump between throughout the day. That activity gets turned into a searchable timeline, and ChatGPT or its coding assistant Codex can use it as memory when you ask them questions later.
This is not OpenAI's first attempt at this idea. An earlier version called Chronicle took actual screenshots of your screen every so often. Computer History drops the screenshots and instead reads raw interaction events through macOS's built in accessibility system, the same technology that powers screen readers for visually impaired users. OpenAI says this means no pictures, no screen recordings, and no audio get captured, and nothing happens in private browsing mode.
The idea of software that remembers everything you do on your computer is not new. A startup called Rewind pioneered this exact concept back in 2022, recording screen and audio to build a personal search engine of your own activity. Microsoft followed with a much bigger and much more controversial version called Recall in 2024, which took constant screenshots on Windows computers and triggered a wave of criticism from security researchers who found it stored sensitive data in ways that were easy to extract. OpenAI's decision to avoid screenshots entirely looks like a direct response to that backlash.
But avoiding screenshots does not remove the risk, it just changes its shape. The memory files that Computer History creates are stored as plain text on your Mac, and they are not encrypted. That means any other program running under your same user account could potentially open and read them. OpenAI also states plainly in its own documentation that the feature raises the chance of prompt injection, where hidden instructions buried in a webpage or document trick the AI into doing something the user never asked for.
For a business, this changes the calculus around whether to allow the feature at all. In Business and Enterprise accounts, a company administrator has to turn on eligibility first, and then each employee has to separately agree to activate it and turn on ChatGPT's memory feature. Users can exclude specific apps or sites, and pause recording whenever they want. OpenAI itself recommends keeping the feature away from apps that handle health records, financial data, or private conversations unless everyone involved has agreed to be recorded.
This puts the decision squarely on whoever manages software policy at a company. Enabling Computer History could genuinely save employees time on repetitive digital tasks, since the automation feature turns spotted patterns into one click workflows. But it also means creating a new file on every employee's laptop that records their behavior in detail and is not locked down. Any company running ChatGPT Business or Enterprise should treat this the same way it would treat new monitoring software: decide on a policy before employees start turning it on individually. The feature is also not yet available in the European Union, United Kingdom, or Switzerland, which suggests OpenAI is still working through how this fits with European privacy rules.