One in 5 firms hit by security incident from shadow AI

Unapproved AI connections have already caused a security incident at 20.1% of firms — roughly one in five — according to DNSFilter's latest survey of 400 IT and cybersecurity professionals across North America. The finding lands despite AI security and governance now ranking as employers' top cybersecurity budget priority, cited by 40.3% of respondents.

The numbers point to a visibility problem inside people-process decisions as much as inside IT. While 54% of firms already use AI governance or monitoring controls to detect incidents involving third-party tools, 46.1% have still had an employee connect an unapproved AI tool to Google Workspace, Microsoft 365 or an SSO environment. One in five of those firms suffered a security incident as a result.

Who is approving AI access?

The permission workflow is where governance breaks down. Among organizations whose employees use AI tools for work, 36.8% allow staff themselves to approve OAuth or API access to corporate systems when a new AI tool requests it — with IT notified only afterward. Another 6.5% grant permission with no notification at all.

"That is 43% of organizations where the decision to connect an AI tool to corporate data is made by someone who does not own the risk," the report read.

For HR leaders, the data reframes an onboarding and policy question: AI tool adoption is effectively being decided at the employee level in nearly half of firms, before IT or security teams see it.

Budgets are growing — but spending is not governance

Employers are responding with money. DNSFilter found 74.5% of firms will have a larger cybersecurity budget than 12 months ago, and 18.8% of those grew their budgets by more than 10%.

AI security and governance topped the priority list at 40.3%, nearly 15 percentage points ahead of cloud security, which 25.5% of respondents named. That gap is the clearest signal yet of where compliance and risk dollars are heading.

The report cautions, though, that governance spending is not the same as governance. A budget can buy a policy, it noted, but tools are still getting connected to the corporate environment — sometimes without visibility.

The visibility gap in numbers

DNSFilter CEO Ken Carnesi quantified the disconnect with traffic data from the company's own network: "On a normal business day, roughly one in seven organizations on our network connects to more than 10 AI tools, yet fewer than one in 100 IT leaders believe that's happening in their environment."

That means actual AI tool sprawl runs at more than 100 times the rate IT leaders estimate. In a 1,000-employee frame, the difference between perception and reality spans dozens of unmanaged connections per day.

"In too many companies, the decision to connect an AI tool to corporate data is being made by someone who doesn't own the risk. Closing that gap has to be the focus of the next round of spending," Carnesi said in a statement.

What about data loss?

The security incidents tracked in the survey echo earlier warnings from browser security firm Menlo Security, which flagged growing workplace shadow AI as a risk for both data loss and data leakage.

"While data loss is a legitimate concern in the enterprise, data leakage, in which sensitive information is inadvertently exposed, can be a bigger issue, particularly when it comes to GenAI," Menlo Security's report read.

That distinction matters for people-process design. Data leakage through generative AI tools often happens without malicious intent — employees paste sensitive material into assistants that were never vetted, a scenario monitoring controls catch only after the fact.

The survey's structure suggests the next round of employer action will target approval workflows rather than blanket bans. With 43% of firms letting non-risk-owners decide on AI connections and budget growth now the norm, expect governance frameworks to shift toward pre-approval requirements and continuous discovery of connected AI tools.