- Chronicle records the macOS screen continuously, with background AI agents converting those recordings into Markdown memory files stored locally on the user’s device.
- Raw screen recordings are deleted after six hours, but the generated memory summaries persist on-device and are stored without encryption.
- The feature is restricted to ChatGPT Pro subscribers on macOS and is unavailable in the EU, UK, and Switzerland.
- OpenAI’s own release documentation warns that Chronicle accelerates rate limit consumption and increases exposure to prompt injection attacks.
What Happened
OpenAI has rolled out Chronicle, an opt-in feature for its Codex coding assistant that uses continuous macOS screen recording to build persistent memory of a user’s active projects, tools, and workflows. As reported by The Decoder, background AI agents convert those recordings into Markdown summaries stored locally, allowing Codex to reference prior context without requiring users to re-explain their environment at the start of each session. Chronicle is currently available as an opt-in preview for ChatGPT Pro subscribers on macOS.
Why It Matters
Chronicle enters a product category that attracted significant scrutiny when Microsoft introduced Recall for Copilot+ PCs in mid-2024. Recall used periodic screen snapshots to build a searchable local memory; it was delayed after security researchers demonstrated that its unencrypted on-device database could be extracted by malware. Chronicle’s architecture — locally stored Markdown files generated from screen recordings, held without encryption — shares structural similarities with that earlier design. OpenAI has geo-restricted Chronicle to exclude the EU, UK, and Switzerland, markets where AI data practices face heightened review under the EU AI Act and equivalent data protection frameworks.
Technical Details
Chronicle operates as a background process, capturing screen recordings that are processed by AI agents responsible for summarizing content into persistent Markdown memory files. OpenAI states that the raw recordings are deleted after six hours, though the derived summaries remain on the device indefinitely. In its own preview documentation, as relayed by The Decoder, OpenAI warns that Chronicle “burns through rate limits quickly, raises the risk of prompt injection attacks — malicious instructions smuggled in through displayed websites — and stores memories unencrypted on the device.” Prompt injection is a known attack class in which instructions embedded in on-screen content, such as a page a developer is browsing, are interpreted as legitimate commands by the AI agent processing that content.
Who’s Affected
Chronicle is currently limited to ChatGPT Pro subscribers running Codex on macOS. Developers managing multiple concurrent projects stand to benefit most from reduced context-setup overhead, but also carry the greatest exposure to the prompt injection risk OpenAI has disclosed, given the volume and variety of content their screens may display. Enterprise users and developers in the EU, UK, and Switzerland are excluded from the preview entirely, with no announced timeline for availability in those regions.
What’s Next
OpenAI describes Chronicle as a preview, with no stated timeline for general availability or expansion to additional platforms such as Windows. The deliberate exclusion of EU, UK, and Swiss markets signals that regulatory engagement or additional privacy review will precede any broader rollout in those jurisdictions. Independent security review of Chronicle’s memory storage model and agent behavior is likely to follow, given the active research community that examined Microsoft Recall under comparable circumstances.
