ChatGPT's "computer history" tracks your clicks and keystrokes: convenience with a trailer

The Verge drew attention to ChatGPT's "computer history" feature: so that the AI agent can pick up the user's work — resume filling a form, return to an interrupted scenario — the system records their actions at the computer, including clicks and keystrokes. From an engineering standpoint it's logical: the agent needs context. From a privacy standpoint, it is the most detailed log of on-screen human behavior mass software has ever collected.

The questions are obvious: where the log is stored, who can access it, whether it feeds model training, and what happens if the account is hacked. In a corporate environment a legal layer is added: recordings of clicks and keystrokes inevitably capture passwords, client data, bank details and trade secrets — everything an employee touches through the working day.

Banning agentic features outright means giving up their real benefits; letting things drift means accepting an unaccounted risk. A working policy for a company in Kazakhstan looks like this: such modes are enabled deliberately, not by default; on machines with access to finance and personal data — only by a separate decision; employees know exactly what is being recorded. And the general principle of the agent era: the more context you give AI, the more useful it is — and the more important it becomes to understand where that context flows.