ChatGPT can now remember what you did on your Mac — without screenshots

OpenAI has begun rolling out Computer History, an opt-in feature in the ChatGPT desktop app for Mac, available to Pro, Business, and Enterprise users. It builds a searchable timeline of activity across permitted apps and websites so ChatGPT can answer questions like "what was I debugging yesterday" without the user re-explaining context.

The piece frames this as a deliberate departure from two prior approaches: Microsoft's Recall, which drew heavy criticism for relying on constant screenshots, and OpenAI's own earlier Chronicle experiment, which used screen captures and OCR. Computer History instead records interaction events: clicks, typing, app switches, and context exposed through macOS's accessibility system. Users can exclude individual apps and sites, pause collection from the menu bar, and the feature is off by default. Data is described as staying local, though the article notes the model itself still gets access to summaries of that activity.

Bias rating

Some bias One or two perspectives missing, some framing that tilts the narrative

What's missing · 01

Concrete technical and security detail on the attack surface created by relying on macOS accessibility APIs, which can expose keystrokes, focus events, and other context, versus pure local processing, plus any independent analysis of whether interaction-event streams could still leak sensitive content via the summaries sent to the model.

What's missing · 02

Enterprise or IT administrator and compliance perspectives, such as data-governance, audit, or regulated-industry concerns, beyond the article's brief mention that employers might prefer the feature stay off.

What's missing · 03

Privacy researchers, security experts, or user-advocacy voices, or even comparative data from Microsoft Recall's fallout, discussing residual risks of long-term activity timelines, potential for workplace surveillance, or the difficulty of truly verifying that only local summaries leave the device.

The other side

While Computer History is opt-in, local-first, and avoids screenshots, it still creates a persistent, AI-readable log of a user's digital behavior that can be summarized and fed into a cloud model. Accessibility-level event capture can reveal more than users realize, including passwords typed in permitted apps, internal document titles, chat thread context, or browsing patterns, and any summary that leaves the device becomes part of OpenAI's ecosystem. In workplace settings this risks turning a helpful assistant into an always-on productivity monitor that managers or auditors could later query, and the "you control which apps" promise is only as strong as the user's diligence and OpenAI's implementation of exclusions. Past incidents with similar "helpful memory" features have shown that convenience features often expand in scope and that users underestimate cumulative privacy exposure; without independent audits or stronger technical guarantees that raw events never leave the machine, many security-conscious individuals and organizations will treat this as a non-starter.