Computer History isn’t really a productivity feature — it’s a data-sourcing strategy dressed as one. By gating it to Pro, Business, and Enterprise subscribers only, OpenAI is recruiting the people already paying it the most money to also hand over the thing web-scraped text can’t provide: minute-by-minute evidence of how professionals actually complete tasks, which is exactly the training-data category every frontier lab is short on as public text runs dry.
Restricting the feature to paid tiers shows OpenAI treating premium seats as a data-acquisition channel first and a convenience feature second.
The mechanics matter for pricing this data. Lifehacker reports OpenAI says it doesn’t retain the raw event files used to build memories, but chat content drawing on those memories can still be swept into training if a user has opted into ChatGPT data-sharing — meaning the real training-data harvest happens one layer downstream, where it’s harder to audit and easier to disclaim. That’s a useful template for other buyers watching this space: don’t pay for the raw log, structure consent so the derivative artifact becomes trainable instead.
The competitive tell is the security corner OpenAI cut to ship fast. PCMag and Thurrott both flag that the interaction-event files are unencrypted and retained locally for 48 hours, accessible to other apps on the same Mac account — a liability Microsoft already absorbed reputationally with Recall’s screenshot approach, and one OpenAI is betting it can avoid by swapping images for structured “events.” Whether regulators buy that distinction is still open: the feature’s absence from the EEA, UK, and Switzerland at launch, noted by both 9to5Mac and Thurrott, suggests OpenAI itself isn’t confident yet. Watch whether Microsoft, Google, or Anthropic answer with their own opt-in behavioral-telemetry features rather than ceding this workflow-data lane to OpenAI and Codex alone.
ChatGPT’s desktop app on macOS has a new feature called Computer History that turns your actions into training data, learning how you work, suggesting automations, and even picking up tasks you left half done.
