AI agent data security for business: who sees what, whose accounts, what support opens
Four questions to ask before AI agents touch company data: who in your team sees their results, whose sign-ins they use, what the vendor's staff can open, and what is logged.
Part of the guide AI employees
Good AI agent data security for business rests on four answers: who inside your company can see what an agent finds, whose accounts it signs in with, what the vendor's own staff can open, and where all of that is written down. In xMap Flows, results follow the visibility of the flow that found them, a person's own sign-in is never shared with the team, xMap support opens an organization only while one of its admins allows it, and every look and change is logged for your admins.
This is for the owner, IT lead or operations manager who has to say yes before a team starts building AI agents on company data. It walks through the four questions, the settings behind each answer, a short setup checklist, and what the product does not claim. For what the agents themselves do, see AI employees.
Question one: who in your team sees what an agent finds?
An agent's results can be sensitive: a list of target accounts, the notes a sales rep keeps, a chat with a customer. In xMap Flows a flow is private, viewable by the team or editable by the team, and can be shared with named people or sharing groups. Results, maps, reports and dashboards follow the flows a person can see.
- Roles: admins, members and viewers. Viewers see but change nothing.
- Custom roles switch off deploying, sending for real, public links, new connections or deleting for some people.
- Owner-only flows: a private flow that uses its owner's personal account is seen by its owner and the people it is shared with only. Admins can pause, archive or delete it without opening it, and its results stay out of their lists.
Question two: whose accounts does the agent use?
Agents send emails, post to Slack, read sheets and write to databases, so each one signs in somewhere. The risk is a colleague's flow quietly using your mailbox.
| Kind of connection | Who adds it | Which flows may use it |
|---|---|---|
| Personal | Anyone | Only flows owned by the person who added it |
| Team | Admins, after a confirmation that says what everyone can do | Every flow in the organization |
| A person's own sign-in to their mail, calendar or files | That person | Always personal; it cannot be made a team connection |
Sign-in details are encrypted and never shown again after saving.
Team connections are kept for accounts that are not one person, such as the Slack bot, a CRM, a database or a service account. The person a team connection signs in as is told the first time a colleague's flow sends with it, and a Slack connection can be limited to the channels flows may post in.
Question three: what can the vendor's staff open?
This is the question vendors most often answer vaguely. In xMap Flows, xMap's own staff get no access to a customer's organization by default. One of your admins grants it under Settings, xMap support access, for between one hour and seven days, and any admin can end it early.
- Support asks, giving a reason; your admins are emailed.
- An admin allows it for a set time.
- While allowed, every page support opens and every change it makes is listed on that card and in Activity, and your admins are emailed.
- Support cannot make lasting access while inside: no new keys, invitations, role changes, shares or public links.
- When the time ends, access ends.
Owner-only flows stay closed to support as they are to your own admins. What Ask xMap keeps for a person, such as notes about them, reminders and what they sent on WhatsApp, is seen by that person only.
Question four: what is written down?
- The activity log records who changed which flow, connection or setting, and admins can download it as a CSV.
- Each run keeps every item each block passed on, so you can see what an agent read and sent.
- An AI agent's run shows how it answered: each tool it called, with what, and what came back.
- Support's visits have their own list, as above.
A setup checklist
- Require two-step sign-in for everyone on the Team page, or turn on single sign-on with your identity provider.
- Set security rules if you need them: allowed network addresses, how long a sign-in lasts, a minimum password length.
- Give people the lowest role that does the job; make a custom role for those who build but should not send.
- Add company systems as team connections; leave personal mail and files as personal.
- Put an Approval step before anything that writes to customers, at least at first.
- On any public Chat page, check the Protection settings: topics, refusing attempts to change its rules, and no personal data in answers.
- Give each AI employee a monthly budget, so a mistake cannot run up cost.
A made-up example
A property management firm in Cape Town lets its leasing team build AI agents. Each agent that drafts tenant emails runs on the leasing manager's own Outlook connection, so it is owner-only: the finance admin can pause it but never reads the drafts. The firm's CRM is a team connection, and its Slack connection is limited to two channels. When a run misbehaved, the operations lead allowed xMap support in for four hours; afterwards the Activity list showed the three pages support opened and the one setting it changed.
What to check honestly
The app, database and stored files are hosted in the United States today, and a choice of region is not offered. xMap does not claim SOC 2 or ISO 27001 certification; the Security page says what exists and what does not. If your policies need either, raise it on a call before you build.
Questions
Can an admin read everything my AI agents find?
Not everything. A private flow that uses its owner's personal account is closed to admins: they can pause or delete it but not open it or see its results.
Can xMap staff look at our flows?
Only while one of your admins allows support access, for a set time between one hour and seven days. Every look is listed for your admins.
Can a colleague's agent send email from my mailbox?
No. A personal connection can only be used by flows you own, and a sign-in to your own mail or files can never be turned into a team connection.
Where is our data stored?
The app, database and stored files are hosted in the United States today. A choice of region is not offered.
Can we require single sign-on?
Yes. An admin can set up single sign-on for your email domain and require it, along with two-step sign-in and other security rules.
Templates that do this
Read next
More on ai employees and automation · Every article · Templates