AI is putting enterprise data under a new kind of pressure. At the very moment organisations are trying to extract more value from their data, they’re also concentrating more of that data into shared repositories, knowledge bases and AI pipelines than ever before.

As AI moves from experimentation into production, datasets that were once separated by business function, sensitivity or operational use are increasingly being brought together to train models and support real-time decision-making.

That shift changes the way organisations need to think about their data infrastructure. What was once primarily a platform for storing and recovering information is increasingly becoming the point at which AI data is consolidated, governed and accessed, bringing new considerations for resilience, compliance and security.

Much of the discussion around AI infrastructure still focuses on models and compute. Those are important considerations, but they can obscure a more fundamental question: how well prepared is the data foundation that supports them?

Here are five ways AI is reshaping enterprise storage security.

1. AI brings data together in new ways

Training a foundation model, or fine-tuning one, typically means pulling intellectual property, regulated customer data, internal knowledge and content such as text, documents, photos, audio and video into a single store that’s open to queries. That is precisely the kind of concentrated target that lets attackers escalate quickly once they gain a foothold. Controlling that risk starts before the data is even stored. Segmenting training data and anonymising or removing sensitive inputs where possible limits the exposure that this kind of aggregation creates, and locking datasets with immutable versioning reduces the risk of quiet tampering.

2. Retrieval-Augmented Generation makes storage an active participant, rather than a passive archive

Connect a large language model to an enterprise knowledge base, and storage stops being something users access directly. It becomes part of every interaction the AI system makes. A misconfigured access control on a RAG index can surface sensitive information through a completely valid query, which is what makes auditability essential. Logging what data is accessed and surfaced supports investigation and compliance, but more importantly, it prevents the kind of structural exposure that’s easy to miss until it’s already happened.

3. Inference workloads introduce a speed problem that manual oversight can’t solve

Production systems such as AI agents or fraud detection engines depend on continuous, low-latency access to data. The pipelines that make this possible operate at a pace that makes manual monitoring impractical, in much the same way that lateral movement across a network can go unnoticed for days if the right controls aren’t in place. Protecting these systems requires securing data in transit, enforcing runtime access controls and ensuring storage systems remain resilient enough that a security incident does not become an operational one.

4. Visibility is the gap that most organisations haven’t closed

Dell Technologies’ Innovation Catalyst research found that 82% of IT decision-makers recognise data as the differentiator for AI integration, and that it must be protected accordingly, yet only one in three say they can turn that data into real-time insight. Much of that gap traces back to the storage layer, and it can be the difference between rapid containment and a days-long blind spot. Without closing it, governance frameworks and policy controls are operating on incomplete information, however well they’re designed on paper.

5. Recovery needs to be rebuilt around AI dependencies

The question organisations should be asking is whether their AI systems can resume trusted operations quickly. That means recovering the correct datasets and model versions, rebuilding data pipelines, and testing recovery against real workloads rather than generic system back-ups. A technically successful restore that brings back the wrong model version, or a broken pipeline, is still an outage in every way that matters to the business.

The practical implication

Storage has become a primary control plane for AI risk. The fundamentals of good security, visibility, immutability, segmentation, Zero Trust and resilient recovery, haven’t changed. What has changed is where and how specifically they need to be applied, and UK regulators are increasingly expecting exactly that specificity; the government’s decision to add ‘digital resilience failure’ to the National Risk Register in July 2026 is a signal of how seriously this is now being taken at a national level.

Organisations that understand their AI data flows, secure storage accordingly, and align controls to the workloads running against it will be better positioned to reduce their exposure, meet regulatory expectations, and build justified confidence in their AI systems as adoption continues to scale.

Stewart Hunwick is field CTO, storage platforms and solutions at Dell Technologies.

Read more

Small Language Models (SLMs) as the gold standard for trust in AI – Stephen Edginton of Dext explains why Small Language Models (SLMs) should be the next stage in your evolving AI strategy

Are you really ready for AI? Exposing shadow tools in your organisation – Without clear AI governance, employees often misuse publicly accessible external tools, known as ‘shadow AI’

Goodbye Software as a Service, Hello AI as a Service – AI as a Service (AIaaS) may replace Software as a Service (SaaS) in the near future. Here’s what that means for your organisation