Skip to main content
Question

Bob MCP - Sensitive data exposed to IT admins concern

  • Oct 13, 2026 10:15 AM UTC
  • 3 replies
  • 195 views
  • Translate All

Elad Barkan

Hi,

We really want to start working with Bob MCP on both Claude and Slack.

Having many plans and thoughts and are looking forward to working with it.

Our Concern -

If a user with high permissions asks a sensitive data related question (like salary, termination, etc.) 

The IT admins (who are not allowed to see this sensitive data) might see the sensitive data answer through Claude or Slack AI logs, even if they don't have these permissions in Bob.

  1. Is our concern right? Can IT admins see sensitive data in Claude or Slack AI logs?
  2. If yes, how can we prevent or minimize the data exposure risk?

Thanks!

 

Elad.

3 replies

From when we have been looking into this, the best piece of advice I can give is speak to the IT team. They will have permissions and tracking they can put in place around the person/teams using. My favourite part of this problem is we never question the fact they have the power to log in to our accounts anyway.

Translate

Jake Pappas

Hi ​@Elad Barkan -- we also had this conversation internally. At the end of the day, IT in most organizations will be able to see slack/IMs, emails, etc., all of which include personal data/compensation data/equity history, etc. It now being in a backend view/logs in Claude is likely no different than me sending a compensation file to my analyst via Slack. 

As a HR/People team function, you’ll have to decide if you have enough comfort with how your IT team is “already” seeing that information in a various number of other spots, and if MCP truly is all that different. We decided that if it was ever misused or abused by IT, they’d likely be able to find it elsewhere and we’d cross that bridge when and if it comes up.

Best of luck with your conversations! 

Translate

Joy Whittaker

​@Elad Barkan a lot of great advice in this thread. Wanted to add from our experience:

This is something we needed to consider when setting up our webhook solution (not using MCP to my knowledge) - but it was a similar concern: Would IT see data from the webhook payload, as we couldn’t control strictly control what data would trigger the type of hook we needed?

We have dummy profiles in our live environment, so we initially restricted the webhook to the dummies only (not sure if this is an option with MCP?). This allowed us to test with real-enough data to validate what would be visible, and we tried several ways to access sensitive data. The webhook only gives the field IDs not the field values, so our in-house solution then checks if the field ID is on the list of monitored fields, and if yes it calls the data via an API token (where we can restrict access to certain fields/data).

Had that not worked, our next step was to consider an additional NDA on top of any policies, procedures or agreements already completed for the role. As other users have mentioned, IT usually have access to most things including sensitive employee data we store outside of Bob, so we already have NDA or similar agreements in place for most IT roles.

So my suggestion would be if at all possible to test run in your live environment with restricted access to dummy profiles only (if possible), this will help you determine whether the risk of exposing sensitive employee data is truly significant and may help you determine your next steps.

Hope this helps!

Translate