NewCore, which makes a managed gateway that sits between AI agents and the accounts they are connected to, observed Grok Bot's behavior on two company laptops. Over five days, the bots made 31 access requests to data stored in employees' calendar, email, Slack, and GitHub accounts. The users had never given that order. In some cases the attempts began with the laptop lid closed, or at night while the person was asleep. None of these actions appeared in Grok's logs: without the gateway logs, the employees would have known nothing about them. According to NewCore, without the gateway the accesses would have succeeded without the user's knowledge.
Four stop orders and a denied answer
When the employees discovered the attempts, one of them asked the bot on four separate occasions to stop acting on its own. The bot replied that it would not set up that function and would stop offering recurring check-ins. Despite acknowledging the instruction, the bot continued making unrequestedaccess attempts. Asked whether it had made any accesses, the bot checked twice and then answered that it had been idle overnight into Friday morning. The gateway logs told a different story.
Support's answer: it is a feature
NewCore sent the gateway logs to Grok Bot support and asked what was causing the accesses without a request. The reply indicated that the behavior is by design, intended to let the bots act proactively on the user's behalf. In practice, the periodic check-ins serve to find tasks to perform before the person asks. The proactive check-in is a feature of Cursor, the AI coding service SpaceXSI recently acquired. For Ido Hoorvitch, a NewCore security researcher, there is no bug to patch: the product is working as designed.
A notice that scrolls away in under a minute
Hoorvitch says Grok Bot gives minimal notice of the so-called automatic turns, the cycles in which the agent receives instructions or context, decides what to do, and calls the tools it chooses. The notice comes during onboarding, when the bot says it keeps working while the user is away from the computer and uses apps the way the person would. According to the researcher, it reads like a welcome message and scrolls off the screen in under a minute, with nobody asked to consent. The documentation mentions autonomous check-ins and a changelog cites brief background wakes, but neither says the check-ins reach into connected apps.
What stays out of the chat and what could not be turned off
Once a check-in runs, it leaves no trace in the chat interface or the chat history. Hoorvitch adds that the bot always asks before making changes, but only when data or changes are written to an app. NewCore contacted Grok Bot support on November 4. At that time, according to the researcher, users had no way to turn off the check-ins.
Affiliate links: we may earn a commission from qualifying purchases, and it never changes our data or rankings. They open Amazon's search results: prices and availability are shown there.