Your AI Already Knows More About Your Business Than You Think It Should

Modern Engineering Solutions

Engineer Your Freedom

The Modern PE
Aug 2026

Your AI already knows more about your business than you think it should.

And you do not know where all of it ends up.

Most firms have not locked down the first problem. Almost nobody has asked about the second.

A company we benchmarked with had someone ask a public AI model a detailed question about their own flagship product.

The answer came back more specific than what their own territory managers knew. Configuration details. Pricing structure. At some point, somebody had queried a public AI tool about that product, the model scraped what it could find, and everything it learned got baked into a system anyone on earth can now ask. Not maliciously. That is simply what a public model does. It learns, and it does not ask permission to remember.

That is the version everyone tells after it already happened. I want to tell you the version that happens before, because what leaks is the advantage you spent years building.

---

The Two Problems

Problem 01

Where your information goes when someone types it into an AI tool

Every prompt, every client detail, every pricing conversation leaves your walls and lands on infrastructure you do not own.

Closed at MES

Problem 02

Who can see it once it is inside your own system

Access governed by job role, not by client or data sensitivity. Nothing yet inspects an answer before it surfaces.

In progress

They get fixed in that order. Fixing the first does nothing for the second.

---

What We Found In Our Own Build

6

Internal AI tools
with no auth check

0

Data retention
setting at MES

$25K

What a competitor
spent on consultants

$0

What it cost us
to fix it

During a security review of our own build, we found that six of MES's internal AI tools, including the one connected to real financial data, had no authentication check at all. Anyone who found the right link could have opened a conversation with an agent that had access to our numbers.

As far as we can tell, nobody outside the company ever did. But the door had been open for a stretch of time nobody had noticed, because it had never been tested, only assumed.

You cannot audit a promise. You can audit a setting.

We set our AI provider's data retention to zero ourselves, inside our own cloud account, and routed every AI call in the company through that single gate. If a client asks where their data went, we show them the setting, not a policy document.

---

Where We Actually Stand

The entire AI layer at MES now runs inside our own cloud account, not the provider's public service. We went into the settings ourselves and set data retention to zero. Every model we use routes through that single locked-down gate. That is the first problem, closed.

A sealed system cannot see the web, so we added a search tool that pulls outside information in. The firm from my opening story built the same tool and caught it quietly opening a two-way connection, data flowing out as easily as in. They cut the outbound half in the code. Ours runs the same way. Information comes in, nothing goes out.

At MES today, your job role decides what you can see. The client and the sensitivity of the data do not. Yet.

That is the honest limit of what is shipped. Most firms have not locked down the first problem. Almost nobody has asked about the second.

Most of what protects you here costs attention, not money.

---

The Question Worth Asking This Week

If you are building AI into your firm, ask your team one question. When someone types a client's name into your AI tool, where does that question actually go, and who controls what happens to it after.

If nobody in the room can answer with certainty, the first problem is still open and the second has not come up yet.

Reply and tell me what you find out.
I read every one.

Michael Groselle, P.E.

Founder and CEO, Modern Engineering Solutions

mod-eng.com  |  contact an engineer  |  214-833-6748