Codex for Mac from OpenAI now takes screenshots and sends them to company servers

OpenAI has added a new Chronicle feature to Codex for Mac — a research preview that gives the AI assistant passive context about what the user is doing on their computer. The mechanism is straightforward: Chronicle periodically takes screenshots, sends them to OpenAI servers for processing, and then saves text summaries locally as unencrypted Markdown files. Essentially, the feature works like a constant observer of the user's screen.

The saved summaries become context for Codex — the assistant can answer questions taking into account what the user did previously: which files they opened, which applications they worked in, what they read. This allows AI to provide more relevant answers without requiring manual explanation of the situation. However, the mechanics raise privacy concerns.

Screenshots first go to OpenAI servers — the company processes them on its side — and only then do text excerpts appear locally. The files themselves are not encrypted, meaning potential access by anyone who has access to the machine. OpenAI has not yet disclosed details about how long images are stored on servers and their deletion policy.

Chronicle's geographic availability is limited: the feature is unavailable to users from the European Union, Great Britain, and Switzerland — apparently due to strict GDPR requirements and similar regulatory regimes in these regions. Additionally, Chronicle requires a subscription costing from $100 per month, which immediately narrows the audience to professional users and developers. This is not the first attempt by a major technology company to create an AI assistant with continuous context through screen capture.

Microsoft announced Recall as part of Copilot+ PC back in 2024, but the feature faced a wave of criticism on security grounds and has been delayed several times. Apple is developing its own tools as part of Apple Intelligence, trying to keep processing on-device rather than in the cloud. OpenAI is taking a more straightforward path: data goes to the server, and the company openly says so.

Chronicle is marked as a research preview — meaning the product is rough and will change. But the direction is clear: AI assistants are moving toward a model of continuous passive observation, where context accumulates automatically rather than being formed by the user manually. The question of how broadly users will accept such a compromise between convenience and privacy is exactly what Chronicle is designed to test.