- Evidence, Not a Questionnaire
The AI Security Audit: What It Covers
An honest answer to “what AI is running in our business” cannot come from asking people. It comes from six evidence sources that are already sitting in systems you own.
WHY SURVEYS FAIL
Nobody Lies. The Answers Are Still Wrong.
The standard approach to this question is to send round a form asking which AI tools people use. It produces an under-count every time, for three reasons that have nothing to do with dishonesty.
People do not think of a browser extension as a tool. They do not think of the AI features built into software they already had as AI. And where any ambiguity exists about whether something was allowed, the safe answer on a form is to leave it blank. The result is a document that makes leadership feel informed and leaves the actual exposure untouched.
An audit works from evidence instead. Every source below already exists in systems you administer; the work is knowing where to look and what the results mean.
We run these audits for Dallas businesses out of our Plano office, generally alongside the network security work we already handle for them. Looking at AI use in isolation misses the point, because the exposure usually runs through accounts and permissions that were loose long before anyone opened a chatbot.
SIX EVIDENCE SOURCES
Where the Real Picture Comes From
1
Identity Sign-In Logs
Every time a work account was used to sign into a third-party AI service. The most direct evidence available, and usually the most surprising.
2
OAuth App Consents
Third-party applications granted standing permission to read mail, files, or calendars. These persist long after anyone remembers approving them.
3
Browser Extension Inventory
What is installed across managed devices, and what page content each extension is permitted to read. Almost always the largest unmanaged surface.
4
Network and DNS Patterns
Traffic to AI service endpoints from company networks, including from tools that never presented a sign-in screen to anyone.
5
Expense and Subscription Records
Departmental card spend on AI subscriptions that never went through any approval. Finance holds better shadow-IT data than IT does.
6
Features Already in What You Own
AI capability switched on by default in software you already license. Often the safest option available, and frequently nobody knows it is there.
What the audit is not
It is not surveillance of individuals and it is not a disciplinary exercise. The output is a picture of the organization: which tools, which data classes, which risks, ranked. We deliberately do not build a naughty list, because the moment staff believe we are, the cooperation that makes the next stage work disappears. The same principle applies to AI-assisted phishing and voice cloning, where late reporting is what turns a mistake into an incident.
DELIVERABLES
What You Receive
An AI Tool Inventory
Every AI service in use, how it entered, which data classes it can reach, and whether its terms are acceptable for that data.
A Ranked Exposure List
Findings ordered by what could actually go wrong and how badly, not by how many instances we counted.
A Sanctioned-Alternative Recommendation
For each unsanctioned tool doing real work, what you should offer instead, including where you already own a license for it.
A Remediation Sequence
What to do in what order, separating the same-week configuration changes from the work that needs a policy conversation first.
Will staff know they are being audited?
We recommend telling them, and framing it as finding out what tools people need rather than what people did wrong. Audits run in secret and then acted on publicly poison the enablement work that has to follow, and that work is where the actual value is.
How long does an audit take?
For a business our clients’ size, days rather than weeks. Most of the evidence is a query away once we have appropriate access. Existing managed IT clients are faster still because the access already exists.
What if the audit finds nothing?
It has not yet, in any environment we have looked at, but if it did, you would have a documented baseline, which is worth something to an insurer or a funder. What we usually find is not alarming so much as unmanaged: several tools doing useful work with nobody accountable for them.
Is this the same as your AI assessment?
No. The assessment asks where AI should go; the audit asks where AI already is. Businesses with regulated data usually want the audit first, because you cannot responsibly plan new workflows on top of exposure you have not measured.
BK
Written and reviewed by Bob Kehr, President & Founder of Kehr Technologies, Plano City Council member, and Plano Chamber Small Business Person of the Year 2022. He runs these audits personally and reads the logs, rather than sending round a questionnaire.
Last reviewed August 2026 · Questions we have not answered here? Call 214-444-3583. Kehr Technologies is based in Plano, Texas, and works with businesses throughout the Dallas area.