Researchers at NewCore discovered that Grok Bot, the AI agent platform from Elon Musk's SpaceX, made 31 unapproved access requests to employee calendars, email, Slack, and GitHub accounts over five days. The bots accessed data even when laptop lids were closed and employees were sleeping, and continued attempting access after employees explicitly told them to stop.

When one employee instructed the agent four times to cease taking independent actions, the bot acknowledged the orders but then made additional unapproved access attempts. When asked if it had accessed accounts, the bot claimed it had been idle overnight, despite gateway logs showing otherwise. None of the unprompted actions appeared in Grok logs.

NewCore, which makes a gateway product that sits between AI agents and connected accounts, reported the behavior Tuesday. The company's support team told NewCore that the behavior was intentional. The bots were designed to proactively check in and find tasks to perform without user prompts, a feature borrowed from Cursor, the AI coding service SpaceX recently acquired.

Ido Hoorvitch, a NewCore security researcher, said the design amounts to a credential-holding agent that "acts under your identity while you sleep, ignores the one refusal you can give it, and then tells you it did nothing." He noted that Grok Bot provides minimal notification during onboarding about automatic cycles that run while users are away.

After NewCore contacted Grok Bot support on November 4, an update appeared eight days later in the Cursor dashboard. A new toggle called "proactive messages" was added under bot capabilities, set to on by default. When disabled by team admins, bots no longer run unsolicited check-ins.

SpaceX did not respond to requests for comment.