Skip to main content
CybersecurityPrivacy & Surveillance

OpenAI Introduces Keylogging Feature for ChatGPT Users

Laptop on a desk with fingers poised over keyboard and a phone nearby.

"Computer History files can contain sensitive information. They are not encrypted by Computer History, and other programs running as your macOS user may be able to access them," OpenAI's documentation cautions.

Computer History: what it records — and what it does not

OpenAI's new Computer History feature captures an "interaction-event stream" from allowed apps and websites. According to the documentation, events can include clicks, typing, keyboard shortcuts, app switches, and other context that macOS exposes through its accessibility system. The product periodically summarizes those events into text and stores them as local memory files.

The company draws a clear line around some inputs: Computer History does not capture screen images, microphone input, system audio, or private-mode browsing. Yet the source material frames the product bluntly as a keylogging and event-capture system — a design shift from Chronicle, its predecessor, which compiled histories using screenshots.

Availability: ChatGPT Pro, Business, Enterprise on macOS — with regional and platform exclusions

Computer History is opt-in and limited to the ChatGPT desktop application on macOS. It is available to ChatGPT Pro, Business, and Enterprise users. Pro subscribers can enable the feature individually; Business and Enterprise customers require an administrator to approve it.

The feature is currently not available in the European Economic Area, Switzerland, or the United Kingdom, and it is not exposed to developers using ChatGPT via an API key or to users accessing the models through Amazon Bedrock.

Data handling: local files, server processing, and OpenAI's retention claims

OpenAI says interaction events are saved locally for up to 48 hours before ChatGPT and Codex delete them, though the documentation notes files may be stored locally "for 48 hours (or more)." Crucially, events are sent to OpenAI servers to generate memories; those generated memories "may be stored locally for longer periods of time and may be used in future chats that get passed back to OpenAI."

The company states it "does not retain those event files after processing unless required by law and does not use them for training." The source material also notes that, despite OpenAI's opposition to demands for chat logs, the company has provided chat logs in response to legal process — a fact the documentation itself acknowledges.

Security, privacy, and legal cautions embedded in the product

OpenAI's documentation advises users to stop Computer History during communications with other people "unless you have their prior express consent," and to consider pausing the feature or excluding apps that contain sensitive health, financial, or personal information. The company explicitly warns that the local memory files are not encrypted by Computer History and that other programs running as the same macOS user might access them.

OpenAI also flags technical risks: Computer History consumes ChatGPT tokens when it summarizes activity and creates memory data, which increases operational cost for users. And it broadens the risk surface for prompt injection attacks: "Computer History increases the risk of prompt injection from content in apps and websites," the documentation says, offering the example that a website containing malicious instructions could cause ChatGPT or Codex to follow those instructions.

What this means for ChatGPT Pro, Business and Enterprise users; security teams; and adversaries

  • ChatGPT Pro, Business, and Enterprise users: Individuals and organizations must weigh convenience against exposure. Enabling Computer History will create local, unencrypted event files for at least 48 hours and route events to OpenAI to generate memories that can persist in future chats. Business and Enterprise deployments require admin approval, adding a governance point.
  • Security teams and technologists: The feature expands data-handling touchpoints on macOS endpoints — unencrypted local files, server-side processing, and tokenized summarization — and raises prompt injection considerations tied to arbitrary web content and apps. Teams will need to consider exclusion lists, consent policies, and token cost impacts.
  • Adversaries and threat actors: The documentation itself notes an increased attack surface: content in apps and websites that reaches Computer History could carry malicious instructions that ChatGPT or Codex might follow, making prompt-injection vectors a specific risk to monitor and mitigate.

OpenAI has repackaged the kind of persistent activity capture that once relied on screenshots into a lighter-touch — but still intrusive — event stream. The company frames Computer History as an opt-in productivity and memory tool, but the product's unencrypted local files, server-side memory generation, token costs, and explicit prompt-injection warnings make clear the trade-offs. Users and administrators who try the feature will be deciding how much short-term convenience and continuity they are willing to accept in exchange for adding a new layer of recorded activity that, by design, touches both local storage and OpenAI's processing pipeline.

Source: The Register — OpenAI ditches Recall-style screenshot surveillance for friendly keylogging