A new study released today puts a number on something most business leaders already feel: AI agents are moving faster than the guardrails around them, and almost every large organisation has already paid the price.

The research, titled “Power without control: Rethinking cybersecurity for the age of agentic AI,” was produced by Economist Enterprise as part of its Tech Frontiers initiative and supported by Rubrik. Researchers surveyed 804 senior business decision-makers at organisations with at least USD 500 million in annual revenue across Australia, France, Germany, India, Italy, Japan, Spain, the UK, and the US.

The headline finding is stark: 98% of organisations have already experienced a disruptive agent-related incident. Nine in ten expect more to happen regardless of whatever safeguards they put in place.

Deploying Faster Than You Can Govern

The central tension the report identifies is a familiar one in enterprise technology adoption, but it’s playing out at an unusual speed with AI agents.

90% of senior leaders say they are deploying agents faster than their security teams can evaluate or govern them. The reason they give is competitive pressure. No one wants to be the organisation that waited while rivals automated.

The problem is that AI agents are not passive tools. They act, make decisions, execute workflows, and increasingly hold access to sensitive systems and data. When they go wrong, they can go wrong at machine speed.

This is where the “power without control” framing comes in. Agents have the power to act autonomously across business systems. The control infrastructure that should accompany that power simply has not kept pace.

The Visibility Gap

Two thirds of organisations surveyed said they cannot tell you what their agents were doing five minutes ago.

That sounds like a process failure, but the researchers frame it as a fundamental architecture problem. Most AI agents were not built with full audit trails, rollback capabilities, or real-time behavioural monitoring as requirements. They were built to be useful, and the governance layer was supposed to catch up.

Only 30% of organisations have robust, tested capabilities to roll back harmful agent actions once they occur. Which means when an agent does something it should not, the majority of businesses have limited ability to contain or reverse the damage quickly.

The Budget Mismatch

Despite near-universal experience of agent-related incidents, the majority of cybersecurity budgets are still pointed at prevention rather than response. The current split is approximately 55% prevention to 45% response and recovery. Organisations expect this imbalance to persist until at least 2030.

The report argues this is the wrong allocation for a world where agentic incidents are already routine. If almost every organisation has had one, and nine in ten expect more, then the question stops being “how do we prevent incidents” and starts being “how fast can we detect, contain, and recover from them.”

What This Means for Business

If you are a business owner or executive deploying AI agents, this research is a useful mirror. The pattern it describes is not unusual — it may describe your organisation right now.

There are a few practical implications worth drawing out:

Governance needs to be designed in, not bolted on. Asking your security team to evaluate an agent that is already operating inside your systems is too late. The design review, access controls, and monitoring approach should be part of the deployment plan, not an afterthought.

You need to know what your agents are doing. If two thirds of large enterprises cannot answer that question in real time, smaller organisations deploying agents through off-the-shelf platforms or API integrations face an even bigger visibility gap. Basic logging and audit trails are table stakes.

Recovery capability matters as much as prevention. The 30% figure for robust rollback capability is a planning benchmark worth taking seriously. If an agent sends incorrect emails to 5,000 customers, or escalates a pricing error across your entire product catalogue, what is your recovery playbook?

Competitive pressure is real, but it is not a governance strategy. The study shows that organisations are using competitive urgency to justify moving faster than their security teams can handle. That is understandable in a market moving this quickly. But it is a risk transfer, not a risk reduction — the incidents are happening at the same rate regardless.

The data from this study confirms what many practitioners already knew: agentic AI deployment has effectively outrun enterprise governance everywhere, not just in a few laggard organisations. That does not make it acceptable. It makes the governance problem more urgent.

---

Enterprise DNA works with businesses across industries to build AI agent workflows that are designed for accountability, not just speed. Explore Omni Ops if you want to talk through how to deploy agents without creating new security blind spots.