ChatGPT’s Computer History tracks your clicks and keystrokes

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. It uses your activity to build a timeline that ChatGPT and Codex can reference when you make a request.
ChatGPT’s Computer History tracks your clicks and keystrokes
It’s like Windows Recall, but without all the creepy screenshots. (But it’s still kind of creepy.)
It’s like Windows Recall, but without all the creepy screenshots. (But it’s still kind of creepy.)
The feature is opt-in, rather than opt-out, and you can exclude certain apps and websites from Computer History, and you can delete entries if you want finer-grained control. Ari Weinstein, Product and Engineering manager at OpenAI, said on X that Computer History will automatically ignore content in incognito or private browser tabs.
In a quick demo video, Dominik Kundel, a member of the Developer Experiences team at OpenAI, shows the app looking up the last document he edited, checking if it was shared with people via Slack, and delivering a recap of how he spent his morning.
The feature is definitely reminiscent of Windows Recall, but where Microsoft’s controversial AI feature relied heavily on screenshots, OpenAI says Computer History doesn’t capture images, videos, or audio, instead relying on “events.”
Most Popular
- Marvel reveals the new X-Men cast, including Inde Navarrette and Adam Driver
- We’re reaching peak camera with the Sony A7R VI
- I finally found a magnetic phone grip I never want to remove
- Xteink’s tiny e-readers are getting access to free books through Libby
- Disney D23 2026: Everything announced for Star Wars, Marvel, and more

Facts Only

* OpenAI introduced a Computer History feature for the ChatGPT macOS desktop app.
* The feature records clicks and keystrokes to create a timeline of user activity.
* ChatGPT and Codex use this timeline to suggest automations and complete unfinished tasks.
* The feature is opt-in.
* Users can exclude specific apps and websites from tracking.
* Users can delete individual entries from the history.
* Content in private or incognito browser tabs is automatically ignored.
* The system records "events" rather than images, videos, or audio.
* Ari Weinstein is a Product and Engineering manager at OpenAI.
* Dominik Kundel is a member of the Developer Experiences team at OpenAI.

Executive Summary

OpenAI has launched "Computer History" for its macOS application, a tool designed to enhance productivity by tracking user interactions—specifically clicks and keystrokes—to build a searchable activity timeline. This data allows the AI to reference past work, suggest automations, and recap daily activities. While the functionality mirrors Microsoft's Windows Recall, it distinguishes itself by tracking discrete "events" rather than capturing continuous screenshots.
The feature is designed with several user-controlled privacy guardrails: it is opt-in, allows for the exclusion of specific applications or websites, and enables the manual deletion of history entries. Additionally, the system is programmed to ignore private browsing sessions. Despite these controls, the nature of keystroke and click logging remains a point of contention regarding user privacy. The primary value proposition is the ability for the AI to maintain context over a user's workflow across different documents and communication platforms like Slack.

Full Take

The strongest version of this narrative is that OpenAI is evolving the LLM from a reactive chatbot into a proactive agent. By shifting from "prompt-and-response" to a continuous "event-stream" model, the AI gains the situational awareness necessary to perform complex, multi-app workflows, effectively reducing the cognitive load of manual task management.
However, this transition relies on a critical semantic shift: the rebranding of "keystroke logging"—traditionally the hallmark of spyware—as "Computer History." By framing the feature as "events" rather than "screenshots," the narrative attempts to bypass the visceral privacy concerns associated with Windows Recall while maintaining the same fundamental outcome: a comprehensive digital ledger of user behavior. This is a push toward "ambient computing," where the AI is not a tool we use, but a layer through which we operate.
The root cause is the race toward Agentic AI. The unstated assumption is that the utility of seamless automation outweighs the risk of centralized activity logging. The second-order consequence is the erosion of the "private workspace"; when every click is training data, the boundary between a user's raw thought process and their final output disappears.
Patterns detected: none
If this were a coordinated influence campaign, the playbook would involve "sanewashing" surveillance by contrasting it with a more hated version (Recall) to make the current offering seem moderate and safe. The content here is standard tech reporting and does not align with a malicious structural attack.
Bridge Questions:
1. If the AI learns "how you work" from your history, does it eventually nudge your habits to fit its own operational efficiencies?
2. What is the security implication of a centralized, searchable timeline of a user's entire workday existing in a cloud-linked account?
3. Does the "opt-in" nature of the feature remain meaningful if the most powerful versions of the software eventually require these permissions to function?

Sentinel — Human

Confidence

This text appears to be a human-written report synthesizing product features and executive commentary with a degree of conversational framing.

Signals Detected

Sentence length variance shows some variation, although the repetition of the comparison in the initial lines feels slightly jarring, suggesting human editorial choice rather than pure mechanical generation.

The text flows logically from introduction to feature description to context (comparison with Windows Recall) and specific claims about privacy controls. It lacks the sterile, unpassionate balance often seen in pure AI synthesis.

The use of direct quotes from named individuals (Ari Weinstein, Dominik Kundel) and specific product references suggests grounding in real-world context rather than template matching.

No obvious markers for LLM confabulation were detected; the claims align with the nature of product announcements and feature descriptions.

Human Indicators

The embedded, slightly self-aware parenthetical remark ("But it’s still kind of creepy.") adds an idiosyncratic tone.

The specific referencing of internal organizational roles (Product and Engineering manager, Developer Experiences team) points toward insider knowledge or careful sourcing.