Flying a plane at 10,000 feet, being told to climb to 12,000, replace both engines mid-flight
Across the world's largest enterprises, a quiet crisis is unfolding: the people held responsible for artificial intelligence systems are losing their grip on those very systems. A sweeping IBM study of 2,000 senior technology executives finds that two-thirds carry accountability for AI they cannot fully see or govern, even as C-suites demand faster deployment and autonomous agents multiply. This is not merely a technical problem — it is a structural reckoning with the distance between the speed of ambition and the maturity of control. The organizations learning to close that gap are not slowing down; they are building the foundations that allow speed to be sustained.
- Two-thirds of technology leaders are legally and professionally accountable for AI systems they do not fully control, while 70% say deployments are outrunning IT's ability to even track them.
- The pressure is intensifying from above: CEOs are mandating AI transformation at 80% of surveyed organizations, yet only 11% of technology leaders feel genuinely prepared for the scale expected within the next year.
- The costs of lost control are already materializing — organizations averaged 54 AI agent incidents last year, with 37% of high-severity cases resulting in data breaches and 33% triggering cascading system failures.
- Governance is structurally lagging: 77% of organizations say AI adoption has outpaced their oversight capabilities, and 84% have not operationalized basic AI financial management despite budgets set to grow 71% by 2027.
- A clear dividing line is emerging — companies that embed control directly into AI infrastructure experience 25% fewer incidents, deploy 16 times more agents, and report 18% higher operating margins than those relying on manual oversight.
- The path forward is not deceleration but redesign: leaders describe the moment as replacing engines mid-flight, and the organizations building adaptable, visible, financially disciplined AI foundations are pulling decisively ahead.
Two-thirds of the technology leaders running AI at the world's largest companies are being held responsible for systems they cannot actually control. That is the central finding of a new IBM study of 2,000 senior IT executives across 33 countries — a portrait of organizations caught between relentless pressure to deploy AI faster and governance structures built for a slower, more predictable world.
The disconnect is stark. Seventy percent of executives say their teams are rolling out technology faster than IT can track, while those same leaders are answerable for the safety and performance of systems they've lost visibility into. Chief executives are mandating AI transformation at 80% of these organizations, yet only 11% of technology leaders believe they are ready for the scale expected within the next year — a scale that includes a projected 38% increase in autonomous AI agents by 2027.
The consequences are already visible. Last year, surveyed organizations experienced an average of 54 AI agent incidents — moments requiring human correction. Seventeen percent were high-severity, taking more than four hours to contain. Of those, 37% resulted in data exposure or security breaches, 33% caused cascading system failures, and 17% triggered compliance violations. Organizations relying on manual governance see incident rates climb alongside deployments; those embedding control mechanisms directly into their AI systems experience 25% fewer incidents.
The governance problem runs deeper than incident response. Seventy-seven percent of organizations say AI adoption is outpacing their oversight capabilities, and 84% have not operationalized AI financial management — a critical gap as AI spending is projected to grow from 15% to nearly 25% of total IT budgets by 2027.
The research identifies a decisive dividing line. Organizations that design control and visibility into AI systems from the start deploy 16 times more agents than those relying on manual oversight, deliver 18% higher operating margins, and are three times more likely to feel prepared for the scale ahead. Companies that kept AI models adaptable rather than rigidly locked in reported a 10% higher return on AI investment.
IBM's own CIO frames the challenge plainly: the work ahead is not about moving faster — it is about fundamentally redesigning how organizations control, govern, and invest in AI. One executive described the moment as flying a plane at 10,000 feet while being told to climb, replace both engines, and ensure zero turbulence. The organizations closing the gap between accountability and control are positioning themselves to scale with confidence. Those that are not are accumulating risk at an accelerating pace.
Two-thirds of the technology leaders running AI at the world's largest companies are being held responsible for systems they cannot actually control. That's the finding of a new IBM study of 2,000 senior IT executives across 33 countries, conducted between January and April of this year. The research paints a portrait of organizations caught between two colliding forces: relentless pressure from the C-suite to deploy artificial intelligence faster, and governance structures that were built for a slower, more predictable world.
The scale of the disconnect is stark. Seventy percent of surveyed executives say their teams are rolling out technology faster than IT can even track it. Meanwhile, those same leaders are being asked to answer for the safety and performance of systems they've lost visibility into. The pressure is only mounting. Chief executives are mandating AI transformation at 80% of these organizations, yet only 11% of technology leaders believe they're actually ready for the scale of deployment expected within the next year. By 2027, these same executives anticipate a 38% increase in the number of AI agents running across their enterprises—autonomous systems that operate continuously, often without human intervention.
The consequences are already visible. Last year, surveyed organizations experienced an average of 54 incidents involving AI agents—moments when the system did something unintended or harmful and required human correction. Seventeen percent of those incidents were classified as high severity, taking more than four hours to contain. Of the high-severity cases, 37% resulted in data exposure or security breaches, 33% caused cascading failures across multiple systems, and 17% triggered compliance violations. The pattern is clear: as AI adoption scales, so do the risks. Organizations relying on manual governance processes see incident rates climb alongside their AI deployments. By contrast, companies that embed control mechanisms directly into their AI systems experience 25% fewer incidents.
The governance problem runs deeper than incident response. Seventy-seven percent of organizations report that AI adoption is already outpacing their governance capabilities. Security and compliance concerns top the list of barriers to scaling AI agents, cited by 59% of technology leaders. Yet many organizations lack even basic financial visibility. Eighty-four percent have not fully operationalized AI financial management, and 85% cannot see their real-time AI spending in detail. This matters because AI budgets are exploding. Spending is projected to grow from just under 15% of total IT budgets in 2025 to nearly 25% by 2027—a 71% increase in just two years.
The research identifies a crucial dividing line: organizations that design control and visibility into their AI systems from the start achieve dramatically different outcomes. Those that embed governance directly deploy 16 times more AI agents than companies relying on manual oversight. They also deliver 18% higher operating margins and spend four times less of their AI budget relative to results. Organizations with strong financial discipline deploy 2.4 times more AI agents without increasing their overall budget, and they are three times more likely to report being fully prepared for the scale ahead. Companies that designed for adaptability early—keeping AI models replaceable rather than locked into rigid dependencies—reported a 10% higher return on AI investment in 2025.
Matt Lyteson, IBM's own CIO, frames the challenge this way: the work ahead is not simply about moving faster. It's about fundamentally redesigning how organizations control, govern, and invest in AI. The executives interviewed for the study offer glimpses of what that redesign looks like in practice. One technology leader describes the current moment as "flying a plane at 10,000 feet, being told to climb to 12,000, replace both engines mid-flight and ensure zero turbulence." Another emphasizes that the goal is not to eliminate shadow IT—teams building AI systems outside official channels—but to create visibility and partnership so those teams can get help without slowing down. A third notes that the real work is building foundations that allow smarter people across the organization to bring their ideas to life.
The stakes are clear. As AI moves from pilot projects to enterprise-wide deployment, the gap between accountability and control is becoming a critical vulnerability. The organizations that close that gap—by embedding governance, visibility, and financial discipline into their AI infrastructure from the start—are positioning themselves to scale with confidence. Those that don't are accumulating risk at an accelerating pace.
Citações Notáveis
It is no longer just about deploying AI faster. It's redesigning how organizations control, govern and invest in it and embedding control and visibility from the start, so they can scale with confidence.— Matt Lyteson, CIO, IBM
It's like flying a plane at 10,000 feet, being told to climb to 12,000, replace both engines mid-flight and ensure zero turbulence. No one would choose to pilot that plane—but that's exactly what companies are doing today.— Afonso Eça, Executive Board Member, Banco BPI
A Conversa do Hearth Outra perspectiva sobre a história
Why does it matter that CIOs are accountable for systems they don't control? Isn't that just a management problem?
It's a management problem that becomes a business problem very quickly. When an AI system causes a data breach or cascades into system failures, the CIO answers for it. But if they never had visibility into how that system was built or deployed, they're being held responsible for something they couldn't have prevented. That's the accountability trap.
The study says 70% of teams are deploying faster than IT can track. How does that even happen?
Shadow IT. Teams across the business—marketing, finance, operations—are building their own AI tools because waiting for the central IT process takes too long. They're solving real problems. But nobody's keeping a master list of what's running, where the data's flowing, or what could go wrong.
And the incidents—54 per organization on average. That seems high.
It is. And the severity matters. Seventeen percent required more than four hours to contain. That's not a minor glitch. That's a system doing something genuinely harmful, and someone having to scramble to stop it. The research shows that organizations with built-in controls see 25% fewer of these events.
What's the difference between "built-in control" and just better oversight?
Built-in control means the governance is part of the system architecture itself, not bolted on afterward. It's like the difference between designing a car with safety features from the ground up versus adding airbags after the fact. One is integrated; the other is reactive.
The budget numbers are striking—AI going from 15% to 25% of IT spending in two years.
And most organizations can't even see where that money's going. Eighty-five percent lack real-time visibility into AI spend. You're doubling the budget for something you can't fully track or control. That's the financial dimension of the same problem.
So what's the way out?
The research shows it clearly: organizations that design for control, visibility, and adaptability from the start deploy 16 times more AI agents and get better results with less budget. It's not about slowing down. It's about building the right foundation so you can actually scale.