Why Boards Are Asking About AI and Data in the Same Breath

There's a question showing up more and more in boardrooms: "What's our AI strategy - and how are we protecting the data that feeds it?"
‍
Not long ago, those were two separate conversations. Today, they're the same one. Here's why that shift is happening, and what it means for security leaders.

The Boardroom Has a New Agenda Item

‍
For years, data security was a CISO problem. The board signed off on budgets and moved on. AI changed that.

Executives watching peers move fast on AI initiatives - deploying co-pilots, agentic workflows, large language model integrations - are realising something uncomfortable: nobody fully understands what these systems can reach.

At a recent executive summit, one CISO opened a session on this topic with a simple statement that landed hard with the room: "I made one of the biggest mistakes of my career by not solving data visibility before we deployed AI. We were building on a foundation we didn't understand."

That's not a rare sentiment anymore. It's becoming consensus.

AI Doesn't Ask Permission. It Just Retrieves.‍

Here's the uncomfortable truth that's driving board-level urgency:

"A RAG pipeline doesn't ask: 'Should everyone really have access to this dataset?' It just retrieves what it can access. That's why visibility and effective access mapping are becoming non-negotiable in the AI era."

AI systems - whether co-pilots, agents, or retrieval pipelines - operate on access, not intent. They don't understand the difference between data that can be accessed and data that should be accessed. That distinction lives entirely in the governance layer. And in most enterprises, that governance layer is years behind.

A joint research study across 2.4 million workers and 3.6 billion permissions found that 96% of enterprise permissions granted to employees are never actually used - and 91% of sensitive data available to workers goes untouched. For human users, unused permissions sit dormant. For an AI agent, that dormant 96% becomes an active attack surface instantly.

Shadow AI Is the New Shadow IT‍
‍

Security teams spent a decade chasing shadow IT - employees using unauthorised SaaS apps. Shadow AI is faster, harder to detect, and carries more risk.

Consider this scenario surfaced at a large organization's quarterly security review: an unauthorised AI tool had been in active use across the company, completely unknown to the internal risk and messaging teams. Nobody had flagged it. Nobody had approved it. It was discovered only because a security platform happened to scan for it.

Meanwhile, separately in the same environment: a single user had a co-pilot tool enabled with access to over 400 million sensitive records.

These aren't edge cases. They are the norm in enterprises that haven't built visibility into what their AI tools can reach.

At a healthcare organization operating under strict compliance requirements, the AI governance posture was described internally as "nascent" - despite having a governance committee in place. The core gap wasn't process. It was observability: security teams couldn't answer basic questions about which AI tools accessed which data, or whether guardrails were actually functioning.

The CISO's Role Is Changing‍

‍

One of the most important shifts happening right now is how the most sophisticated security leaders are reframing their function.

Rather than positioning data security as a control that slows down AI adoption, leading CISOs are making the case that data visibility is what makes AI adoption possible.

The argument goes like this: if you can classify your data, understand who and what can access it, and map the exposure risk before an agent ever touches it, you can say yes to AI initiatives that others are forced to block. Security becomes the enabler, not the gatekeeper.

Industry analysts are beginning to formalise this framing. One analyst recently described data security as increasingly "supplying the policy logic" that other security pillars enforce - meaning it's no longer one layer among many, but the foundation everything else is built on.

The CISOs who are winning board conversations are the ones who walk in with a concrete answer to the question: "Before we deploy this AI system, here's exactly what data it can reach, who else can reach it, and what we've done to close the gaps."‍

Agentic AI: A New Risk Category‍

‍

If co-pilots raise the stakes, agentic AI raises them again.

A co-pilot answers questions. An agent takes actions - it queries databases, calls APIs, sends communications, modifies records. It operates continuously, at machine speed, without the natural hesitations and judgment calls a human brings to the same tasks.

Security teams at organizations actively deploying agentic architectures are asking questions that weren't on the radar eighteen months ago:

• "Can we distinguish between an agent and the data it has access to?"
• "If a user has legitimate access to PII for one task and uses an email agent for another, how do we prevent the two from being bridged?"
• "What happens when an AI lab runs on a developer's laptop and we have no visibility into what it's connecting to?"
  ‍‍

These aren't theoretical. They are active concerns in the security programs of large enterprises across financial services, healthcare, real estate, and technology.

At a working session with a large organization, a discovery exercise revealed an external AI-connected identity with access to millions of sensitive records - including social security numbers, credit card data, passport information, and bank details. There was no clear ownership. No approval trail. No governance documentation. The reaction from the security team: "I sent this to my management team right away."

That moment - the unexpected discovery of real, material exposure - is what's driving urgency at the board level. It's no longer abstract.

The Question Boards Are Now Asking‍

‍

The shift we're seeing is this: boards are no longer asking "Are we secure?" They're asking "Can we deploy AI responsibly - and how do we know?"

Those are fundamentally data questions. And they require fundamentally different answers than traditional security reviews can provide.

The organizations moving fastest on AI - and doing it with confidence - are the ones that built data visibility first. They know what data they hold, where it lives, who and what can access it, and what the exposure looks like before any AI system is switched on.

That's not a nice-to-have. In the age of agentic AI, it's the prerequisite for everything else.

AI isn't creating data chaos. It's exposing the chaos that already existed.

The board is asking about AI and data in the same breath because they're the same problem. The organizations that recognise that early will be the ones that get to say yes.

‍Where Do You Start?
‍

Knowing there's a problem is one thing. Knowing where you stand is another. Here's how organizations are beginning to get concrete answers.

Go deeper with the AI Security Readiness Assessment. For organizations that need a more comprehensive, expert-led evaluation, this assessment evaluates your organization's AI security posture across eight critical domains - covering AI governance, data security, infrastructure, model risk, supply chain, application security, operations, and compliance. The assessment produces a detailed maturity baseline, evidence-backed gap analysis, and a clear, prioritised roadmap for deploying AI safely and responsibly.  Get started today.

‍