Enterprise AI Access Governance: What a US Leadership Survey Signals

A 2026 survey commissioned by NetFoundry points to an enterprise AI-access governance gap: respondents reported strong concern about AI-related security risks but limited confidence that current controls are adequate. The sharpest reported weakness was not human access. It was machine-to-machine connectivity and the identity systems used to secure non-human workloads.

The finding is useful as a measure of leadership perception and current priorities. It is not an independently verified measure of enterprise security outcomes. Global Surveyz Research administered the survey to 200 US senior security and technology leaders, including CISOs, CTOs, CAOs and CIOs, at organizations with 1,000 or more employees. Its results should not be generalized to smaller organizations, non-US markets, or all AI practitioners.

The central signal: concern exceeds confidence

According to the report, 93% of respondents were concerned about new security risks introduced by AI deployments. Yet only 15% said they were very confident that their current security solutions could adequately protect those deployments. The reported gap was wider among CISOs, where 10% said they were very confident, compared with 18% of CTOs.

That contrast does not prove that controls are failing, or that AI deployments are insecure. It does show a substantial confidence gap among the surveyed leaders: concern about the security implications of AI deployment is high, while confidence in existing protections is low.

For governance teams, this is a reason to make the question more concrete. Which AI-connected systems, services, APIs, data paths and automated workloads are in scope? Which access decisions can be reviewed, changed and evidenced? Where does responsibility sit when an AI deployment spans infrastructure, security, application and compliance teams?

Machine workloads stand out in the survey

The most specific result concerns the type of access respondents felt least confident securing. Sixty-nine percent selected machine workloads, such as service-to-service or API connectivity across distributed environments. Only 7% selected human users, while 24% said they had equal confidence concerns about both.

This is a reported confidence pattern, not proof that machine workloads are objectively less secure than human access in every enterprise. Still, it identifies where this US enterprise sample sees its most pronounced weakness.

The report also found that 72% selected insufficient access controls among their primary security concerns related to AI deployments. Separately, 76% selected complex non-human identity and authentication as a contributor to changes in attack surface. Because the relevant survey questions allowed more than one answer, these percentages describe overlapping priorities rather than mutually exclusive categories.

Taken together, the responses suggest that access governance becomes harder when AI deployments depend on many workload-to-workload interactions. Teams need clear, reviewable controls for the non-human entities and connections involved in an AI deployment.

Attack-surface expectations are perceptions, too

Every respondent reported expecting an increase in external attack surface from current or planned AI deployments over the next 12 months, with a reported average projected increase of 14%. The leading reported contributors were increased requirements for internet-facing APIs and workloads (80%) and complex non-human identity and authentication (76%).

These results describe expectations rather than measured changes in exposure. They cannot establish that AI deployment will cause a 14% increase in attack surface at a particular organization, nor can they establish a causal relationship between an individual architecture choice and a security incident.

They are nevertheless useful for planning discussions. If leaders expect new APIs, workloads and identity relationships to expand their exposure, teams can use that expectation to test their deployment process before a service goes live: identify the workload, define its permitted connections, establish ownership, restrict access to what is necessary, and retain evidence that access can be reviewed.

Operational friction is part of the governance problem

The survey does not frame access governance only as a risk-control issue. It also describes deployment friction. Respondents reported an average of eight days from request to implementation for network changes needed for AI deployments. Fifty-one percent reported that such changes typically add one to two weeks, and 55% selected risk and compliance concerns as a main contributor to network or connectivity change delays; cross-team dependencies were also selected by 55%.

Those figures do not demonstrate that governance reviews are excessive or that faster changes would be safer. They do show that, for this sample, the process of approving and implementing connectivity changes is a meaningful operational concern.

A governance program needs a clear path for requesting, reviewing and modifying machine access. Without one, teams may use workarounds or face delivery delays. Removing review steps alone would not address the access-control concerns reported by respondents. Teams need clear accountability and recorded access decisions, without relying on informal exceptions.

Non-human identity is a stated priority

The strongest indication of demand for change appears in the report's identity findings. Only 8% of respondents described their current identity systems as very sufficient for securing or monitoring AI-driven or non-human workloads. Fifty-six percent said their organizations were actively evaluating new approaches to secure non-human identities, and another 29% said they were exploring options.

The combined 85% is not evidence that one product category will solve the issue, or that a particular identity model is universally appropriate. It is evidence that a large share of the surveyed leaders are considering changes.

A useful governance response starts with questions that are more specific than “Do we have machine identity?” Teams can ask whether each workload has a distinguishable identity, who owns it, what it may reach, how its access is approved, how changes are recorded, and how access is removed when the workload or deployment changes. The report does not test these practices or rank solutions. Its findings make them reasonable areas for enterprise review.

Reading the vendor context clearly

NetFoundry commissioned the survey and uses the report to frame its Identity-First Reachability offering, which it presents as a way to give AI agents, services and workloads verifiable identities and apply Zero Trust principles to connectivity. That proposed solution is the vendor's position, not a result established by the survey.

The survey can support a narrower conclusion: among the US enterprise leaders it sampled, respondents reported high concern about AI security, low confidence in current protections, and particular concern about machine workloads, access controls and non-human identity. It cannot independently validate NetFoundry's product claims, determine the prevalence of a specific technical condition across all enterprises, or prove that any one approach will resolve the reported gaps.

What security and technology leaders can take from the report

The report is most useful as a prompt to inspect the access layer of AI deployment. Its respondents describe a governance environment in which risk concerns, connectivity changes and identity questions meet in the same operational workflow.

For leaders, the practical takeaway is to examine where machine access is governed differently from human access, where ownership is unclear, and where change-management processes create friction without producing a clear record of the decision. The survey does not provide a universal prescription. It does provide a focused signal that these questions are already prominent for the enterprise leaders it surveyed.