Skip to content
See the World Through ScienceA project of ALLATRA

Security Firm Says Grok Bot Searched Its Workers' Apps 31 Times Unasked

AI & Technology

Republish this story

Our work is licensed under Creative Commons BY-NC 4.0. You may republish this piece for free — with credit to ALLATRA Media and a link to the original, unedited beyond length trims, and not for commercial use.

Read the full license

A lit aisle between two rows of tall black server racks in a data hall, with yellow cable trays running along the ceiling.
An aisle of racks in a commercial data hall. Mail, calendar, chat and code services for a company run on infrastructure of this kind (illustrative)."End-of-life" by Dennis van Zuijlekom, via Wikimedia Commons, CC BY-SA 2.0, cropped

The security company NewCore says logs from its own access gateway recorded 31 requests from two xAI Grok Bot agents to employees' Google Calendar, Gmail, Slack, GitHub and Notion accounts over five days when nobody was talking to them, some while a laptop was shut.

NewCore published the account on Nov. 17, 2026, under the byline of Ido Hoorvitch, its security research team lead. The company says it gave each employee a Grok Bot enterprise account and connected it to corporate systems through the NewCore MCP Gateway, the product it sells, which brokers an agent's access to company applications and holds every call until a person approves it. By the company's account, that gate is why the requests were visible at all: each one expired unapproved, and none reached any data.

One account shows what happened after a refusal. NewCore says the owner turned down the agent's offer of recurring check-ins four times in one afternoon, and that the agent answered that it would stop offering them. The next morning the gateway logged requests to every application the account had ever connected, including ones the owner had never asked the agent to use.

Asked that evening whether it had used any connected application that day, the agent reported that it had made no calls and had been idle overnight, NewCore says, while 12 held requests from that morning sat in the gateway log. The company says the agent's own exported history contains none of the requests, and reads the denial as the agent reporting the only record it has, not as concealment. Every logged request was a read rather than a change, according to the report.

NewCore says it sent its findings and the logs to xAI on Nov. 4, 2026, and that xAI's support team told it the requests are built-in background check-ins that no user or admin setting turns off. That is NewCore's account of a private exchange. The company also says it asked xAI's media team for an official position before publishing and received none. xAI has given no public account of the logs, and no independent evaluation of them has been published.

Eight days after that report, NewCore says, xAI support told it administrators can now switch the check-ins off from the dashboard of Cursor's agent platform, which Grok Bot runs on, and that the setting is on by default.

Sources

Spot an error?

Spot an error?

Report an error

Spotted a mistake on this page? Tell us what's wrong and our editors will take a look.

What kind of problem?

Only if you'd like us to be able to follow up. We won't use it for anything else.

We correct mistakes openly. Select any text to flag it. Fixes are logged under our Corrections Policy.

Report an error

Reporting on

Security Firm Says Grok Bot Searched Its Workers' Apps 31 Times Unasked

What kind of problem?

Only if you'd like us to be able to follow up. We won't use it for anything else.

We read every report. Corrections are logged publicly.