Organizations
are comfortable granting access.

They’re far less prepared to take it back.

227 IT and security leaders reveal what happens when AI agents gain autonomy.

The surprise isn’t how much agents can now do, it’s how little control most organizations have once they’re doing it.

RealVNC technology is trusted by more than 60,000 organizations

2026 Agentic AI Survey

About the survey

RealVNC surveyed 227 IT and technical decision-makers across the US, UK and Canada in September 2026. The findings are broken down by industry, because being in control looks different by sector.

Pages from the RealVNC 2026 Agentic AI Survey eBooks

227 respondents

IT and technical decision-makers surveyed across the US, UK and Canada.

  • Software and IT services bloc of the survey.
  • Healthcare, finance and public sector bloc.
  • Industrial and operational bloc.
Open pages from a RealVNC 2026 Agentic AI Survey eBook

The missing brake

Organizations are comfortable granting access. They’re far less prepared to take it back.

71.8% of organizations already let AI agents act without a human approving it first, and of those, 75.5% have no way to revoke that access once it’s granted. The gap isn’t awareness, most believe they’d notice if an agent overstepped, it’s that the controls to stop it, or even see it happen, often aren’t in place.

Agentic AI: Understanding Your Real Exposure

A practical guide to assessing risk, control, and readiness in an agent-driven world.

Agentic AI is moving faster than most security and IT teams can keep pace with, and most of what you’ll read about it right now is either hype or alarm, neither is that useful when you’re the one who has to decide what your own organization can safely let an agent do. We ran this survey because we kept hearing the same thing from IT and security leaders: not “is AI risky”, they already know that, but “how exposed are we, specifically, and what do we actually do about it”. This report is built to answer that, not to warn you about it.

See where you stand

Compare your own answers against 227 real IT and security leaders who’ve already answered these same questions.

Spot the access you can’t see

Understand how agents are already reaching legacy and operational systems nobody built with this in mind.

Close the confidence gap

Learn why even teams certain they’d catch a rogue agent usually can’t prove it.

Understand the real driver

It isn’t how many controls you have, it’s which ones, and where they sit.

Make the case for the right fix

Not more tooling, a shift from controlling access to controlling action.

Know what to check this week

The report’s own recommended response, not just the risk.

Choose the report for your industry

Three mini-reports break the full survey down by sector. Not sure which one fits? Start with the full report below, every sector is covered there too.

Agentic AI eBook: Software and IT services

Software and IT services

Well resourced on paper, but the ability to stop an agent falls away fast once autonomy becomes routine. If your team has already given agents standing access to production, this is where to check how much control you actually kept. 48 respondents.

Agentic AI eBook: Healthcare, finance and public sector

Healthcare, finance and public sector

Cautious about granting autonomy, but not yet ahead on the ability to back that caution up once access is granted. If your organization operates under compliance oversight, this is the gap your auditor will eventually ask about. 56 respondents.

Agentic AI eBook: Industrial and operational

Industrial and operational

Exposure on legacy and OT systems is rising, and few organizations can stop an agent once it’s in. If an agent can already reach a PLC or an engineering workstation on your floor, this report shows what that access actually looks like elsewhere. 61 respondents.

Organizations are granting AI agents more autonomy than they can comfortably control.

Our 2026 Agentic AI Survey asked 227 IT and technical decision-makers across the US, UK and Canada how far they let AI agents act, what those agents can reach, and what has already gone wrong. The pattern holds across every finding: organizations are comfortable granting access, and far less prepared to take it back.

The full report sets out the findings in detail, including:

  • Why most organizations let agents act without approval, yet few can revoke that access
  • How deeper agent access to legacy and operational systems makes agents harder to stop
  • Why unsanctioned agent use is associated with more incidents
  • Why confidence in detecting a rogue agent often outruns real-time monitoring
  • Why the question is which controls matter, and where they sit

Join our presenters as they go through the complete research findings and hear the panel discuss what the results mean for security, governance and AI adoption.

Neil Gad
Neil GadChief Product and Technology Officer
Nick Cavalancia
Nick Cavalancia4X Microsoft MVP
Anton Zdziebczok
Anton ZdziebczokVP of Product
Watch the webinar: AI Agents Are Acting. Can Anyone Hit the Brakes?

A preview of the data

75.5%

Once access is granted, most organizations can’t take it back.

28.2%

Already happened, not a future risk: an AI agent incident in the past 12 months.

42%

Confidence in catching a rogue agent is outrunning the ability to actually do it.

Source: RealVNC 2026 Agentic AI Survey. Online survey of 227 IT and technical decision-makers in the US, UK and Canada, fielded September 2026. The figures above have different respondent bases; the webinar will explain the findings in context.

Frequently asked questions

If you can’t find an answer to your question, please visit our Help Center or get in touch with us.

Get in touch
How was the survey conducted?

An online survey of 227 IT and technical decision-makers in the US, UK and Canada, fielded in September 2026. The findings are broken down by industry, because being in control looks different by sector.

Which report should I read if my industry isn’t listed?

Answer to come.

Is this free?

Answer to come.

Can I request more than one report?

Answer to come.

Still deciding? Start with the full picture.

AI Agents Are Here. Can Anyone Hit The Brakes? covers all 227 respondents, across every industry, in one report.

Download the full report