OpenAI is testing Computer History in its ChatGPT desktop application for macOS, transforming user keystrokes, clicks, and daily system telemetry into live training ground for task automation. The feature logs a chronological timeline of low-level OS interactions that ChatGPT and Codex query to stitch together fragmented tasks, infer context, and resume unfinished workflows.
To sidestep the security backlash that crippled Microsoft's Windows Recall, OpenAI claims Computer History avoids screen recordings, audio, and visual snapshots entirely. The telemetry operates on an opt-in basis, complete with log deletion tools and app-specific blocklists. As confirmed on X by OpenAI product lead Ari Weinstein, the utility automatically suppresses logging inside private and incognito browser sessions.
During an internal demonstration, OpenAI Developer Experience engineer Dominik Kundel showcased ChatGPT tracking down a user's latest document edits, cross-referencing whether the file had been transmitted via Slack, and compiling an end-of-morning activity digest. While technically impressive for personal productivity, streaming granular interaction logs into a third-party model provider turns corporate workstations into active data harvesters.
For enterprise CISOs and engineering leads, this setup presents an immediate compliance hazard: background event logging inevitably risks funnelling proprietary code pipelines and sensitive business logic into OpenAI's cloud.