Dell has delivered over 6,500 AI Factories to customers across nearly every industry, giving us a front-row view into what separates organizations that scale AI from those that stall. From that vantage point, we see a common barrier to scaling related to the security and resilience strategy for AI solutions. We commissioned independent research from Omdia to validate and expand on what we’re already seeing in the field with organizations leading in AI adoption. This vantage point matters even more right now with AI safeguards taking center stage. As organizations look to capitalize on the benefits of AI, they’re defining their own strategies, deploying sovereign and private AI solutions that keep data inside their own walls to contain risk by design.
The topline finding is one every infrastructure buyer should sit with: the market is moving fast, but governance, controls and investment priorities are still real hurdles to broader adoption. Enterprises aren’t underinvesting in AI security so much as struggling to align governance, ownership and technical safeguards as adoption scales.
The data confirms the wave is real: 98% of organizations are already using or developing agentic AI for IT operations alone. This isn’t a pilot-program trend; it’s the new operating model. The difference between scaling and stalling increasingly comes down to whether security and resilience were built in from day one or bolted on after the fact.
Four gaps holding adoption back
Omdia’s research aligns with our observations in four specific ways, quietly capping how far and how fast organizations can scale AI use cases across their organization.
The first is a governance gap. Nearly a quarter (24%) of organizations still govern agentic AI permissions through manual processes — a model that doesn’t scale alongside autonomous agents acting at machine speed. It shows: 79% of organizations have already experienced an AI-related incident in the past 12 months, and a meaningful share involve agentic-specific failure modes like excessive agency, prompt injection and unpredictable tool usage, not just classic breaches.
The second is a confidence paradox. That same 79% incident rate sits alongside sky-high confidence: 95% say they’re confident or very confident in their ability to manage AI risk. Those two numbers shouldn’t coexist so comfortably. It suggests many organizations are pattern-matching AI risk to traditional IT risk management, when agentic AI actually behaves, fails and needs to be recovered differently.
The third is a policy enforcement gap. 93% of organizations that allow public GenAI use have some usage policy in place — encouraging on its face. But 55% of security leaders believe it’s likely or very likely employees are violating those policies anyway. A policy on paper and a policy that actually shapes behavior are two different things.
The fourth is budget misalignment. Nearly every organization expects to increase AI security spending over the next 12—18 months, yet that spend isn’t converging on a clear priority — no single investment area breaks 40% (the top pick, AI threat detection tools, sits right at that ceiling). Ownership is just as fragmented: CISOs hold sole budget authority for AI security tools in only 22% of organizations, behind the CTO (37%) and IT (26%). Money is flowing in, but without a clear owner or priority, it struggles to translate into a coherent posture.
Two disciplines, one strategy
Cybersecurity and resilience are related but distinct problems. Cybersecurity stops the threats — traditional ones we know well, plus new AI-native ones like prompt injection, model manipulation and agent identity abuse. Resilience is different: making sure that when something does go wrong, backup and recovery are orchestrated and automated enough to quickly restore the AI application, its models and its data to a known-good state.
Organizations are starting to connect these dots. 87% rated threat detection and response (such as MDR) as an effective or very effective technology to secure the use of AI, while backup and recovery tools are rated by 88% of respondents for addressing AI resilience — right alongside traditional threat detection and endpoint tools. Omdia’s analysts called this out directly, noting it illustrates “the convergence of security and resilience.” You can’t call your AI strategy secure if you haven’t also answered the resilience question: if this agent, model or data pipeline is compromised tomorrow, how fast can you bring it back clean?
The pull toward sovereignty and isolation
A third thread runs through the data: control. Nearly 4 in 10 organizations deploy agentic AI in on-premises data centers, and 43% use on-premises edge environments — an appetite for keeping AI infrastructure and data within their own walls that tracks with growing customer demand for sovereign AI solutions and workload isolation, including sandboxing and air-gapping, which roughly a third of responding organizations already use to contain agentic AI risk.
The call to action: start the conversation early
If there’s one takeaway for infrastructure and IT buyers, it’s this: involve your security and resilience teams at the planning stage, not the deployment stage. The organizations winning with AI are the ones who invite security and resilience leaders in before the architecture is locked, closing the governance, policy and budget gaps this research surfaces before they become someone else’s incident.
That’s the posture Dell wants to help every customer take. Security teams shouldn’t be the department of “no” that shows up after the AI Factory is already built and starts blocking rollout. They should be co-architects from the first whiteboard session — helping the business move faster and adopt AI safely, because the plan for stopping threats and the plan for recovering from them were designed together from the start. Read more about our approach to securing AI Factory.
Omdia’s research identifies four gaps: a governance gap (24% of organizations still govern agentic AI permissions manually), a confidence paradox (79% incident rate vs. 95% confidence), a policy enforcement gap (93% have GenAI usage policies, yet 55% of security leaders believe employees violate them anyway) and budget misalignment (no single AI security investment priority exceeds 40%, and CISOs hold sole budget authority in only 22% of organizations).
Dell reported this
Source: www.dell.com
Source link
