What Is Shadow AI, and Is It Already Running Inside Your Series A?

I asked a Series A founder last month how many AI tools his eight-person team was using. He said three: the coding assistant, the support bot, and the internal search tool they had all agreed on in a Slack thread back in March. Then I asked his head of ops the same question. Her list had eleven items on it, including a resume screener nobody remembered approving and a transcription tool that had been quietly summarizing every customer call since February.

Neither of them was lying. They were just answering different questions. He knew what the company had decided to use. She knew what the company was actually using. That gap has a name now, and it showed up with real numbers behind it this month.

The word for the tool nobody approved

Shadow AI is not some exotic new risk category invented by a cybersecurity vendor trying to sell a dashboard. It is the plain fact of employees using AI tools or services without formal approval, oversight, or integration into official company systems, which is exactly how Smarsh and FTI Consulting defined it in their 2026 Enterprise AI Trends Study, published July 7. Swap “AI tools” for “cloud apps” and this is just shadow IT, a problem every company has lived with since the first employee expensed a Dropbox subscription. AI just moved faster and got more useful, so more people started doing it.

The study found that 55% of enterprises are actively deploying AI. Only 26% say their governance frameworks are fully aligned with that pace. Just 30% report having comprehensive capabilities to even detect and manage the AI tools running outside their approved workflows. That is not a company failing at AI. That is nearly three quarters of companies succeeding at adoption faster than they can build the paperwork to match it, which will sound familiar to anyone who has ever shipped a product before writing the support documentation.

What a fifteen-minute inventory actually finds

Here is the approach I would run this week in that founder’s shoes, and it takes less time than a standup.

First, ask two different people the same question I asked that founder: what AI tools does the company use. Ask a founder or exec, then ask whoever is closest to daily operations, support, or sales enablement. The gap between those two lists is the actual shadow AI footprint, and it will not be zero. PagerDuty’s June survey of 1,250 office professionals at $500 million-plus revenue companies found 66% had used AI tools at work despite believing it was against company policy, and 88% had shared work-related information with a public AI system: 43% pasted in emails, 40% shared meeting notes, 34% entered customer data, 31% uploaded financial documents.

Second, resist the urge to react to that list by banning anything. Every operator I talk to who tried an outright ban got the same result: usage went underground instead of away. Wakefield Research’s data backs this up directly. 39% of respondents said they would rather use AI without telling anyone, and that number climbs to 47% at companies with $1 billion or more in revenue. Bigger, more locked-down companies produce more hiding, not less.

Third, sort the list into two columns: tools touching customer data or financial data, and tools that are not. The first column needs a decision this week, either sanctioned with a data-handling agreement or replaced with an approved alternative. The second column can wait for the next quarterly review. This is not about achieving perfect AI governance in an afternoon. It is about finding out where the actual exposure sits.

Why the confident founders are the exposed ones

The part that should give any founder pause is not the usage numbers. It is the confidence numbers. A March study of 292 executives and 492 knowledge workers by Okta, conducted with Apprize360 across seven countries, found that 90% of executives expressed confidence in their visibility into AI tool usage at their own company. Meanwhile 52% of those same companies’ knowledge workers admitted to using unapproved AI tools. The executives were not wrong that a policy existed. They were wrong that the policy was working.

I have seen this exact pattern play out in diligence calls. A founder tells an investor with total sincerity that the company has an AI policy, because it does, somewhere in a wiki page from eight months ago. Nobody checked whether anyone follows it. That is not dishonesty. That is the same overconfidence the Okta study measured at scale, just with a term sheet attached to the outcome this time.

A shadow AI problem does not mean the team is careless. It means nobody wrote down the answer to "which of these are we actually okay with."

The numbers that actually matter at your stage

Forget the aggregate industry statistics for a second and focus on the two numbers that actually determine the exposure. The first is the gap between what leadership thinks is approved and what the team is using, which the Okta data pegs at roughly 38 percentage points between stated confidence and actual unapproved use. The second is the data-sensitivity split inside that gap: PagerDuty found nearly a third of unauthorized AI use involves customer or financial data specifically, not just internal notes. That second number is the one an acquirer’s legal team or a Series B investor’s diligence checklist will actually ask about. A managed IT provider, Magna5, put it plainly in a note to mid-market clients this month: companies are adopting AI faster than they can govern it, and the fix is not fewer tools, it is visibility into which ones are running and what data they touch.

Ship it

Run the two-list exercise this week, not next quarter. The goal is not a zero-shadow-AI company, because that does not exist at any stage. The goal is a company that knows its own list, has sorted it by data sensitivity, and can answer the question honestly the next time someone asks. That answer, more than any policy document, is what actually protects a founder in the room. And honestly, once the real list is on the table, most of it turns out to be completely fine. The relief is usually bigger than the risk.