ChatGPT's Computer History logs what you do on your Mac
ChatGPT Computer History records clicks, typing, and app switches on macOS for Pro, Business, and Enterprise. Off by default — here's the policy to set first.
OpenAI shipped a feature that turns the ChatGPT desktop app into a recorder of your workday. ChatGPT Computer History builds a timeline of what you did on your Mac and hands it to ChatGPT and Codex as context. It is off by default and, on Business and Enterprise plans, an admin has to unlock it first. That admin decision is the whole story — make it deliberately, before someone in your company makes it for you.
What actually happened
Per OpenAI's documentation, Computer History records interaction events — clicks, typing, keyboard shortcuts, and app switches — across apps and websites you allow, using macOS accessibility permissions. It explicitly does not capture screenshots, screen recordings, microphone input, system audio, or private-mode browsing. 9to5Mac notes it replaces Chronicle, an earlier research preview that worked from screenshots.
The storage model is the part worth reading twice. Raw event files sit locally in the app's container and auto-delete after 48 hours. The memories generated from them are plain-text Markdown written to ~/.codex/memories/extensions/skysight/, and they persist until you delete them. The feature does not encrypt them. Clearing history removes both the events and the memories built from them, in 10-minute, hourly, daily, or full increments.
It is macOS-only, limited to Pro, Business, and Enterprise, requires the Memories feature, and needs admin approval under Workspace Settings → Permissions & roles — which grants permission to opt in, not the feature itself. OpenAI's own warning: the feature "can increase prompt-injection risk from malicious content in apps or websites," and it recommends pausing collection during private conversations.
Why it matters for your business
The upside is real. Computer History can spot a repeated workflow and offer to turn it into a reusable skill or a scheduled task — that is SOP capture from actual work instead of from a document nobody wrote. For a small team, that is the cheapest process documentation you will ever get.
The exposure is equally real, and it is not the recording. It is that an unencrypted plain-text file on a laptop now summarizes what your bookkeeper, your ops lead, or your founder did all morning — and that a hostile web page inside an allowed site is now upstream of an agent that can act.
Set three things before you flip the admin switch: an allowlist of apps and sites narrow enough that payroll, banking, and your password manager are never in it; a rule that the memories directory is treated as sensitive on any machine covered by disk encryption and backup policy; and a named owner who reviews what got captured. Then turn it on for one person and read the output before anyone else gets it.
Key takeaways
- Computer History logs clicks, typing, shortcuts, and app switches on macOS — no screenshots, audio, or private browsing
- Raw events auto-delete after 48 hours; generated memories are unencrypted plain-text Markdown that persist until deleted
- macOS only, on Pro, Business, and Enterprise; off by default, and Business/Enterprise admins must grant permission first
- OpenAI warns the feature increases prompt-injection risk from content in allowed apps and websites
- It can convert repeated workflows into skills or scheduled tasks — useful SOP capture, if you scope the allowlist first
Who decides what your staff's AI tools are allowed to watch? We set up AI workflows with explicit scopes, audit trails, and automation that runs on systems you control instead of a laptop's memory file. See how we build automation, or talk through your AI policy.
- #chatgpt
- #ai-governance
- #shadow-ai
- #macos
- #prompt-injection
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Intelligent Terminal 0.2 runs agents on a local model
Microsoft's Intelligent Terminal 0.2 adds bring-your-own-model support, per-tab agent selection and WSL-native execution. A coding agent that never calls a cloud API.
Read itDeepSeek Harness is MIT-licensed. Read the loop.
DeepSeek open-sourced Harness v0.1 under MIT — a plugin-first agent runtime where models, tools, sandboxes and loops all swap in config. The harness is the product.
Read it