OpenAI is moving toward a model of artificial intelligence that does more than answer prompts. Its latest ChatGPT feature, called Computer History.
Gives the assistant the ability to remember aspects of how users interact with their computers, including clicks, typing, keyboard shortcuts and application switching. The development offers a revealing glimpse into the broader direction of OpenAI’s consumer AI ambitions.
Particularly as the company prepares its first hardware device with designer Jony Ive. Computer History is currently an opt-in feature for ChatGPT’s macOS application.
Rather than continuously taking screenshots, as Microsoft’s controversial Recall initially did, OpenAI says its system records interaction events through macOS accessibility functions.
Register for the next Tekedia Mini-MBA.
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Screenshots, screen recordings, microphone input and system audio are not captured, while private browsing activity is excluded. The collected events can be transformed into a searchable timeline that ChatGPT and Codex can use as contextual memory.
The distinction is important because OpenAI is attempting to solve one of the central problems facing personal AI: context. Traditional chatbots depend heavily on information users deliberately provide. An agent that understands what someone has already done, which document they edited.
Which application they used and what task remains unfinished can potentially become much more useful. That capability, however, comes with significant privacy implications. Even without screenshots, a sufficiently detailed record of clicks.
Keystrokes and application activity can reveal sensitive information about a person’s work, communications and habits. Reports have also raised questions about how locally stored activity data is protected and the risks posed by malicious software or prompt injection.
OpenAI has attempted to put controls around the feature. Users can choose whether to activate Computer History, exclude particular applications or websites, pause tracking and delete individual activity records. The feature is therefore materially different from a system that secretly monitors activity.
The broader question remains: how much visibility should an AI assistant have into a person’s digital life? That question becomes even more important when viewed alongside OpenAI’s hardware ambitions.
The company is working with Ive and his design team on its first consumer device, expected to arrive during the second half of 2026. OpenAI has described the partnership as an effort to rethink how people interact with computers, while reports have suggested a screen-free device capable of functioning as a new type of AI computer.
Computer History could represent part of the software foundation for that future. Instead of requiring users to repeatedly explain their activities, an AI assistant could understand ongoing context and intervene when useful.
The ultimate objective is not simply conversation; it is persistent, ambient assistance. But this also creates a delicate trade-off. The more an AI knows about a user, the more capable it becomes—and the greater the consequences if that information is misused, exposed or misunderstood.
OpenAI’s challenge will therefore extend beyond building intelligent hardware. It must convince consumers that an AI capable of understanding their digital lives can also be trusted with them. The success of its future device may depend as much on that balance between intelligence and privacy as on the hardware itself.