OpenAI is rolling out ChatGPT Computer History, an opt-in feature in the ChatGPT app for Mac that records your activity across apps and websites you select, in order to build a searchable timeline. Available to Pro, Business and Enterprise subscribers, it replaces the Chronicle preview and has not yet reached Europe. It ships with real privacy controls, but it carries one serious flaw: the memory files it creates locally are not encrypted.
What ChatGPT Computer History actually records
The feature relies on macOS accessibility APIs, the system interfaces originally designed for screen readers and assistive tools. They let an app observe what happens in other apps, provided the user has granted permission.
In practice, Computer History records interaction events:
- mouse clicks;
- keyboard typing;
- keyboard shortcuts used;
- switching between apps.
These raw events are then periodically converted into text summaries and memory files stored locally, then organised into a timeline sorted by day and time. In a demonstration reported by the press, a member of OpenAI’s developer experience team showed the system locating the last document he had edited, checking whether he had shared it through Slack, then generating a summary of his morning.
OpenAI stresses what the feature does not do: it captures no screenshots, no screen recording, no microphone input and no system audio. Private browsing is also excluded. That is the main claimed difference from Microsoft’s Recall feature, which relied on periodic screenshots and triggered strong controversy.
The real problem: plain-text files on your disk
This is where it falls short. The memory files generated by Computer History are plain-text, unencrypted Markdown files stored on your Mac.
Two retention periods coexist:
- interaction event files stay on the machine for up to 48 hours before deletion;
- the memory files derived from them remain indefinitely and unencrypted, until you clear them manually.
Why does this matter? Because on macOS, any program running under your user session can read files accessible to that session. A malicious extension, a utility downloaded from a dubious site, or an infostealer — malware specialised in harvesting local data — could therefore comb through this history with little effort. It would not need to spy in real time: the summary of your working days is already written and filed.
Worth noting: OpenAI publicly documents this behaviour and issues a warning about it. This is not a hidden discovery but a deliberate design trade-off, on which users must make an informed decision.
Why OpenAI is building this memory anyway
The reasoning is consistent with where the industry is heading. An assistant becomes markedly more useful when it has context: what you have already done, which files you opened, who you communicated with, which tasks remain unfinished. Without memory, every exchange starts from scratch and you spend your time re-explaining your situation.
OpenAI is betting that the value of that continuity justifies the collection, provided usage is bounded. The feature does offer real controls:
- choose precisely which apps are included;
- pause collection at any time;
- delete the recorded history.
OpenAI also states that Computer History uses fewer tokens than the Chronicle preview it replaces, suggesting more economical upstream processing.
What this means for you
If you use ChatGPT on a Mac in a professional setting, a few precautions are in order before enabling the feature.
- Check your company’s legal framework. A detailed activity history may conflict with your data protection obligations, particularly if you handle client information.
- Exclude sensitive apps. Password manager, encrypted messaging, banking tools, medical records: none of these belong in a timeline stored in plain text.
- Enable full-disk encryption with FileVault. It does not protect against malware running during your session, but it does protect if the machine is stolen.
- Purge the history regularly if you do not need long-term memory.
- Remember the feature is opt-in. It is not enabled by default.
For European users, the question does not arise yet: the EU rollout is still pending, most likely because of GDPR requirements around data minimisation and the legal basis for processing.
Towards an AI-maintained activity history
Beyond the ChatGPT case, this announcement outlines a shift in personal computing. Assistants are becoming systems that continuously maintain a record of what you do. Microsoft attempted it with Recall, Apple is advancing system-wide context mechanisms, and OpenAI is proposing its own approach here.
The design challenge is no longer giving an assistant more memory: it is establishing strict boundaries around collection, retention, permissions and deletion. Put differently, an AI-maintained activity history becomes a new layer of the operating system — and therefore a new attack surface to secure.
Key takeaway
ChatGPT Computer History is more privacy-conscious than Recall on one key point — no screenshots — but its plain-text local storage is a genuine weakness. The feature remains optional and configurable: it is up to you to decide whether the context gain is worth the trade-off.
Would you enable this kind of memory on your work machine? Share your view in the comments, and browse our other articles on personal data protection at Wanda-techs.com.
Read also:
Share this content: