OpenAI is rolling out a new feature for ChatGPT’s macOS desktop app called Computer History, which tracks a user’s clicks and keystrokes to build a timeline of their digital activity. The feature is designed to help ChatGPT and its coding companion Codex understand how a person works, offering suggestions for automation and even picking up tasks that were left incomplete. It has drawn immediate comparisons to Microsoft’s controversial Windows Recall, though OpenAI is taking a different approach by avoiding screenshots altogether.
Computer History is opt-in, meaning users must explicitly enable it, and it offers granular controls to exclude specific apps and websites from being tracked. Users can also delete individual entries if they want finer-grained control over what the AI retains. Ari Weinstein, a product and engineering manager at OpenAI, confirmed on X that the feature automatically ignores content in incognito or private browser tabs, adding a layer of privacy for those who want to keep certain browsing sessions off the record.
The feature works by recording “events” — discrete actions such as opening a document, clicking a button, or typing a message — and organizing them into a chronological timeline. When a user later asks ChatGPT for a recap of their morning or wants to pick up where they left off on a project, the model can reference that timeline to provide contextually aware assistance. In a demo video shared by Dominik Kundel, a member of OpenAI’s Developer Experiences team, the app is shown locating the last document he edited, checking whether it had been shared over Slack, and delivering a summary of his early workday activities.
How Computer History Compares to Windows Recall
Microsoft’s Windows Recall, introduced earlier this year, achieved notoriety for taking regular screenshots of a user’s screen every few seconds, running optical character recognition, and storing the resulting text in a searchable database. That design raised significant privacy and security concerns, especially after security researchers found that the data was stored in plaintext on some systems. Microsoft was forced to delay the rollout and make changes, including encrypting the database and requiring biometric authentication before the feature can be used.
OpenAI’s Computer History takes a different path. According to the company, it does not capture images, videos, or audio. Instead, it relies on event-based tracking, which captures metadata about user actions without producing a visual recording of the screen. This approach is less intrusive on its face, but still raises questions about the extent to which an AI system should be allowed to monitor a user’s behavior.
The semantic difference is significant. Screenshots can reveal everything on a screen, including sensitive documents, personal messages, and confidential emails, all of which can be compromised if the database is breached. Events, by contrast, capture only the specific actions a user takes, such as clicking a button or typing a particular string of text. This is more analogous to how many productivity trackers and analytics tools operate, but the AI's ability to interpret those events and make inferences adds a new layer of capability and concern.
The Privacy Trade-Off
Even with the opt-in nature and exclusion controls, Computer History represents a significant expansion of the data that OpenAI can collect from users. The feature is explicitly designed to turn user actions into training data, which means it could be used to improve OpenAI’s models in addition to providing immediate utility. OpenAI has not disclosed exactly how long the timeline data is retained or whether it remains on-device or is uploaded to cloud servers. The company has stated that users can delete entries, but the default behavior and retention policy remain unclear.
Privacy advocates have long warned about the dangers of ubiquitous surveillance software, even when it is optional. The fact that Computer History is opt-in rather than opt-out is a step in the right direction, but the feature still normalizes the idea of an AI assistant watching every move a user makes on their computer. Critics argue that the feature could be used to build detailed psychological profiles, inferring everything from work habits to emotional states based on the timing and nature of certain actions.
There is also the question of security. Any database that records a user’s activities over a long period becomes a high-value target for hackers. Even if OpenAI encrypts the data in transit and at rest, the sheer volume of sensitive information could make it a tempting prize. If a cybercriminal gains access to a Computer History timeline, they could potentially see writing drafts, code snippets, and other intellectual property that the user has interacted with.
OpenAI has, however, built in some safeguards. Incognito and private browser tabs are automatically ignored, and users can manually add specific apps and websites to an exclusion list. This allows users to prevent sensitive services like banking or health care portals from being tracked. But the effectiveness of these controls depends on how rigorously they are enforced and whether the underlying event collection is truly comprehensive.
Potential Use Cases and Benefits
Proponents of Computer History point to the potential for significant productivity gains. By understanding a user’s patterns, ChatGPT could automate repetitive tasks such as formatting emails, generating reports, or organizing files. It could also help users define complex workflows, such as “Every time I download a PDF, save it to my research folder and send a summary to my team.” The timeline feature would allow the AI to see what happens before and after a given action, enabling it to learn which actions are typically followed by others.
The ability to resume unfinished work is another major selling point. If a user closes their laptop for the night and picks it up the next morning, ChatGPT could reconstruct the state of their projects and help them prioritize. In the demo, the app not only found the last document edited but also checked if it had been shared via Slack, demonstrating an understanding of collaboration context. This kind of integrated awareness could make ChatGPT feel like a much more natural assistant, one that remembers not just the words a user types, but how they fit into a broader workflow.
For developers, the integration with Codex is particularly intriguing. Codex is OpenAI’s code generation tool, and with Computer History it could potentially see the entire context of a developer’s activity: which files were opened, which functions were modified, which build errors occurred, and how they were resolved. This could lead to highly contextual coding assistance, where the AI suggests fixes based on the exact sequence of actions the developer just performed, rather than just the contents of a single file.
The Broader AI Context
Computer History is part of a larger industry trend toward context-aware AI systems. Companies like Google, Meta, and Amazon have been experimenting with AI assistants that can access email, calendars, and message history. The difference is that Computer History goes one step further by capturing the user’s actions at the operating system level, rather than just within a single application.
This raises the stakes in the ongoing debate about how much personal data AI companies need to collect in order to deliver useful services. OpenAI has argued that the current generation of AI models, which are largely trained on static datasets, will soon be supplemented by systems that can learn from real-time user behavior. Computer History is an early example of that vision, but it is not without its risks.
There is also a question of consent and transparency. While the feature is opt-in, users may not fully understand what they are signing up for when they enable it. The term “Computer History” is innocuous enough, but the reality is that every click and keystroke is being logged and potentially used to train AI models. OpenAI’s documentation will need to be clear about what is collected, how it is used, and how long it is stored. The company may also need to provide periodic reminders to users that the feature is active, rather than letting it silently run in the background.
Security and Ethical Considerations
OpenAI has a history of dealing with security scares. In March 2023, a flaw in ChatGPT’s open-source library exposed the chat histories of some users, leading to concerns about data privacy. The company has also faced lawsuits over the use of copyrighted material for training. Adding a feature that records clicks and keystrokes could exacerbate these concerns if the data is not handled properly.
One approach that could mitigate some risks is on-device processing. If the computer history timeline were stored and processed locally, with only aggregated or anonymized insights sent to OpenAI’s servers, the risk of a massive data breach would be lower. However, many AI features require cloud-based inference, especially when they use large language models such as GPT-4. It remains to be seen whether OpenAI will adopt a hybrid approach that keeps raw logs local while using cloud APIs for specific tasks.
Another ethical concern is the potential for subconscious self-censorship. If people know their actions are being tracked by an AI, they may alter their behavior in ways that are not entirely conscious. This could have a chilling effect on creativity and exploration, as users become more guarded about the documents they open or the sources they consult. The feature is designed to be helpful, but its presence might change the very workflow it is trying to assist.
What Lies Ahead
As OpenAI continues to roll out Computer History, the response from the public will likely shape how the company refines the feature. If privacy concerns dominate the conversation, OpenAI may need to add more safeguards or provide clearer opt-out mechanisms. If users embrace the feature and find it genuinely useful, similar tools could become standard across many applications and operating systems.
The introduction of Computer History also sets a precedent for future AI systems that are embedded into desktop environments. Microsoft has been pushing Copilot, Google has introduced Gemini features in Chrome, and Apple is working on deeper Siri integration. Each of these companies is likely watching closely to see how OpenAI’s approach is received, and whether the market accepts an AI that actively monitors user behavior.
In the meantime, users who enable Computer History will be able to enjoy the benefits of a more contextually aware assistant, but they should also be mindful of the trade-offs. Regularly reviewing the timeline, excluding sensitive apps, and deleting unnecessary entries can help mitigate privacy risks. OpenAI’s decision to make the feature opt-in and avoid screenshots indicates that the company is aware of the criticism surrounding Windows Recall, but only time will tell whether these measures will be enough to reassure users that their clicks and keystrokes are in safe hands.
Source:The Verge News