Shadow AI at your Mexico operation: employees already use ChatGPT, Gemini and agents with client data (how to set rules without slowing anyone down)

On October 12, 2026, two stories told the same tale from different angles. The Hacker News warned that employees’ personal AI assistants are already logged in to company email, Drive and CRM. And El Economista, one of Mexico’s main business papers, described how unapproved AI ends up creating more work for the business instead of saving it.

That is shadow AI: AI tools people use for work without the company knowing or approving them. Your headquarters may already have an AI policy. The question is whether anyone at the plant or office in Mexico has read it.

Why shadow AI hits Mexico sites harder

In our experience, global policies often arrive in English, sit on an intranet and never get rolled out locally. Meanwhile the payroll clerk pastes a report into ChatGPT for a quick summary. The sales rep connects an assistant to Gmail to reply faster. Both are trying to get the job done.

The trouble comes later. Answers nobody checked go out to a client. A spreadsheet with sensitive data stays in the history of a free account. An assistant stays connected to someone’s mailbox after they leave.

Attackers are getting sharper too. El Sol de México reports that AI is making digital attacks in Mexico harder to detect. Every personal app wired into company email is one more door.

How to find which AI is in use today

Look before you set rules. Four places to check:

• Sign-in logs. Microsoft 365 and Google Workspace record every time someone signs in to an app with their company account.
• OAuth apps. OAuth is the permission behind “Sign in with Google” or “with Microsoft”: the app gets into the account without knowing the password. Here is how to audit those apps.
• Browser extensions. Plenty of writing helpers read everything on screen.
• Firewall traffic. The firewall controls what goes out to the internet; its reports show which AI services get visited and from which machines.

Mexican data that must never leave

Data	Example	Into a personal AI account?
Tax data	RFC (tax ID), digital seal certificates	Never
Payroll	Salaries, CURP (national ID), bank accounts	Never
Invoicing	CFDI e-invoices, XML files	Never
Clients	Names, phones, contracts, pricing	Never
Access	Passwords, keys, system screenshots	Never
Public text	Draft social post, campaign idea	Yes, without names

Personal data of clients and employees is protected under Mexican law, on top of whatever your global privacy program requires.

A one-page AI policy for the local team

One sheet, in Spanish, with this:

1. Approved tools. Which ones, by name, and for what.
2. Business accounts. The business tier, owned by the company, with a contract that states what happens to your data. Read that clause before paying.
3. Forbidden data. The table above, as is.
4. Connections. No assistant gets into email, Drive or the CRM without IT sign-off.
5. Who to ask. A local name and a WhatsApp number, not a global ticket queue.

Our take: banning everything does not work. People keep using AI, just from their personal phones, where you see nothing.

People and data, not agent permissions

This is about what your people do with their own AI. If you are worried about assistants installed on the PCs themselves, see how to control AI assistants on the PCs of your Mexico operation.

The technical piece that closes the loop is DLP, data loss prevention: rules in Microsoft 365 or Google Workspace that spot an RFC, a card number or a CFDI in an email or file, then warn or block before it leaves.

Let’s talk

What about your Mexico operation?

Within a ProcessBi IT support plan we audit apps and access in Microsoft 365 or Google Workspace, set up DLP and help your local team adopt a one-page policy in Spanish, aligned with headquarters.

It starts with a free assessment of what is connected today.

Book your free assessment — we reply the same business day.