A new five-part research series from VentureBeat surveyed 573 enterprise professionals in June 2026 and found a pattern that should concern every business leader deploying AI right now: most organizations think they have control over their AI stack. Most do not.

The report is called “The Control Gap,” and the headline statistic is this: 85% of enterprises run two or more AI platforms, each claiming to be the organisation’s “primary” AI layer. Only 8% have consolidated to one. And only a third — 38% — have a central team governing AI across the business.

The rest are governing by hand, by habit, or not at all.

The Numbers That Matter

The research paints a detailed picture of how enterprise AI governance has fallen behind enterprise AI adoption:

• Only 38% of organisations report a central team governing AI across the business
• 17% say no role holds formal accountability for AI across the stack — not a team, not a person, not a policy
• 32% name “no single accountable owner” as their top barrier to effective governance
• 85% run two or more competing AI platforms claiming to be the primary layer
• 25% have already experienced a runaway agent incident — an autonomous agent running beyond its intended scope and generating unexpected cost or operational impact
• 21% track agent spend only through after-the-fact logs with no real-time kill switch in place
• 49% identify shadow AI — unapproved agentic pipelines running on personal or team accounts outside any oversight — as their most severe control failure

The survey sample skews toward mid-market and lower-large companies (100–2,499 employees), which makes these numbers more striking, not less. These are not early adopters experimenting recklessly. They are established organisations with procurement processes and IT teams — and they still do not know who owns AI.

Why This Is Happening

VentureBeat’s framing is useful: this is an ownership problem, not a technology problem. The platforms exist. The models are capable. The gap is that no one drew an org chart around them.

Several dynamics contributed to this. AI adoption moved through business units faster than IT policy could catch up. Individual teams — marketing, sales, operations, finance — built workflows on whatever tools they could access. Those workflows now exist in production with no central registry, no spend visibility, and often no named owner who would be called if something went wrong.

The 85% platform fragmentation figure is a direct outcome of this. When each team adopts its own AI layer and no central function reconciles them, the result is competing stacks, duplicated spend, and governance that is effectively impossible to enforce because no one can see the whole picture.

The Financial Risk Is Already Showing Up

The 25% runaway agent statistic is the one that should focus attention. One in four enterprises surveyed has already been hit by an autonomous agent that exceeded its intended scope — and the primary mechanism through which this shows up first is billing.

Agents that are poorly scoped can spin into loops. Agents that trigger other agents can cascade. Agents with write access to external tools can take irreversible actions. When the only detection mechanism is an after-the-fact log — as it is for 21% of respondents — the first signal a business gets is an unexpected invoice or a partner flagging an unexpected action.

This is not a hypothetical. It is what one in four enterprise AI deployments has already experienced.

What Good Governance Actually Looks Like

The 38% of organisations with central AI governance are pulling ahead. The research identifies some consistent patterns among them:

A single owner with cross-stack visibility. Not a committee. Not a policy document. A named function — often a Chief AI Officer, an AI Platform team, or an expanded IT governance mandate — that can see every deployed model, every active agent, and every integration point.

A real-time kill switch. Agentic AI that cannot be stopped quickly is a liability. Effective governance means the ability to pause, inspect, or terminate any agent run without waiting for a log review.

An AI inventory. Most organisations cannot list the AI tools their teams are using. Central governance starts with a registry — what is deployed, who owns it, what data it touches, and what it can write to.

Spend control by design. Agentic AI is not a flat monthly fee. It scales with usage and can spike unpredictably. Governance that does not include budget limits, spend alerts, and per-team allocation is governance in name only.

What This Means for Business Leaders

VentureBeat’s research closes a loop with the AI adoption story of the past two years: companies raced to deploy, platforms competed for penetration, and governance was treated as a second-priority problem that could be solved later.

Later has arrived.

The shadow AI finding — 49% naming unauthorised agentic pipelines as their top control failure — is particularly relevant for businesses that have handed teams broad AI access. If your employees are building agent workflows on personal accounts or departmental subscriptions, those workflows are running in production. You likely do not know what they can access, what they have already touched, or what happens if one of them fails.

The implication is not that AI deployment should slow down. It is that governance needs to catch up to the pace of deployment rather than lag behind it. That means establishing ownership, auditing what is already running, and building control infrastructure before the next phase of agent adoption rather than in response to an incident.

---

Enterprise DNA’s advisory work focuses on exactly this problem — helping organisations build the governance layer that makes AI adoption sustainable rather than fragile. If you are navigating AI ownership and control challenges in your business, talk to our team about what a structured approach looks like.