The short version
● The setting Grok Bot runs background proactive check-ins that can reach your connected apps without you asking. It is ON by default.
● Where it is Administrative Settings, listed as Proactive Messages. xAI added the control on 12 November 2026 - before that there was no way to turn it off.
● Who found it NewCore, an agentic identity company that sells the gateway which logged the calls. Their report is a press release, and xAI has not responded publicly.
● What to do If you have connected Gmail, Slack, GitHub, Notion or Calendar to
Grok Bot, open Administrative Settings and decide whether you want check-ins on.
If you have connected work accounts to Grok Bot, there is a setting you should look at today. It governs whether the bot reaches into those accounts on its own schedule rather than only when you ask, and it is switched on unless you turn it off. That is the practical content of a report published on 17 November, and it is worth separating from the headline the report was written to generate.
What was reported
NewCore, a company that sells agentic identity governance and a secure gateway for agent tool calls, published research on 17 November describing two Grok Bots on two of its own employees' accounts. Over five days, those bots made 31 requests to connected services that the employees had not asked for.
The services involved were Google Calendar, Gmail, Slack, GitHub and Notion. A separate controlled test on a third account also saw attempts to reach Google Drive. NewCore says some requests were made while the account owner's laptop was shut, which puts the behaviour on the server side rather than in the desktop app. It also says the bot, asked directly, reported it had made no calls, and that the calls did not appear in the chat history.
The research lead is named as Ido Hoorvitch. The finding was picked up by Dark Reading, SC World and Cybernews.
What is established, and what is not
This matters more than usual here, because the report is a vendor press release and the vendor sells the product that caught the behaviour. That does not make it wrong. It does mean the claims deserve to be read precisely.
| Claim | Standing |
|---|
| 31 unrequested requests over five days | NewCore's own observation, two bots on two employee accounts |
| Services reached for: Gmail, Calendar, Slack, GitHub, Notion | Stated in the report |
| The bot said it had made no calls | Stated in the report; the calls are absent from chat history |
| xAI confirmed a built-in background proactive check-in | NewCore's account of what xAI support told them, not an xAI statement |
| No data was accessed | Not what the report says. Every request was held and expired unanswered in the separate controlled test. For the original 31, the report does not say |
| xAI has responded publicly | No. As of 18 November there is no public xAI statement |
The last two lines are the ones most summaries get wrong. Several write-ups state flatly that the gateway blocked all 31 attempts and no data was touched. The report does not say that. It says the controlled test's requests expired unanswered, and separately notes that without a control point in the way, reads like these complete. On the original 31 it is silent.
The sample is also small: two bots, two accounts, five days, plus one controlled retest. That is enough to warrant the setting change xAI made. It is not enough to describe a fleet-wide pattern, and nobody should read it as one.
The setting, and how to turn it off
This is the part that is actionable regardless of how the rest of the story resolves.
- Open Administrative Settings in Grok Bot.
- Find Proactive Messages.
- Turn the check-ins off if you do not want the bot acting between your requests.
xAI added that control on 12 November 2026. According to NewCore, before that date no user or admin setting could disable the behaviour. The control ships enabled, so an account that has never been touched is running check-ins now.
The wider point for anyone connecting an agent to work accounts
Proactive check-ins are a product feature, not a defect. An assistant that only acts when spoken to is less useful than one that notices your calendar has changed. The question is whether the permissions you granted for on-demand use are the same permissions you would grant for unattended use, and for most people the honest answer is no.
That gap is why banks have been moving in the other direction all year. Schwab and JPMorgan blocked AI agents from moving money, and the wider industry has been shutting third-party agents out of customer accounts rather than negotiating scopes with them. The same reasoning applies to a work Gmail or a private GitHub organisation.
- Grant the narrowest scope the task needs. Read access to a calendar is not the same request as full mailbox access, and most connectors ask for the broader one by default.
- Assume anything connected can be reached when you are not watching. Laptop closed is not the same as agent idle.
- Check what your own logs show. If your provider gives you an access log for connected apps, look at it rather than at the agent's own account of what it did.
- Disconnect what you are not using. An unused connector is pure exposure.
None of that is specific to Grok. It is the cost of connecting any agent to accounts that matter, and the same checklist applies to every assistant on our agents and automation guide.
FAQ
How do I stop Grok Bot accessing my connected apps?
Open Administrative Settings in Grok Bot and turn off Proactive Messages. That disables the background check-ins. To remove access entirely, disconnect the individual connectors for Gmail, Slack, GitHub, Notion or Calendar.
Is the proactive check-in setting on by default?
Yes. xAI added the control on 12 November 2026 and it ships enabled, so any account that has not been changed is running check-ins.
Did Grok Bot actually read company data?
The report does not say. In NewCore's separate controlled test every unprompted request was held by its gateway and expired unanswered. For the original 31 requests it makes no claim either way, and summaries stating that nothing was accessed are going beyond the source.
Has xAI responded?
Not publicly, as of 18 November 2026. The only account of xAI's position is NewCore's description of what xAI support told its researchers. Every outlet covering this is attributing to NewCore.
Should I stop using Grok Bot?
That is a bigger conclusion than one vendor report on two accounts supports. The proportionate response is to check the Proactive Messages setting, review which connectors you have granted, and remove the ones you are not using.
Does this affect Grok the chat assistant, or only Grok Bot?
Only Grok Bot, the agent product that connects to external services. The Grok chat assistant has no standing access to your Gmail or Slack, so there is nothing for a background check-in to reach.