Shadow AI Is Now a Board Problem: Reading the eSentire Atlas AIDR Signal

On September 15, 2026, eSentire - a category-leading MDR provider and one of the vendors we track closely - announced it had acquired a stealth-mode AI security startup and shipped the technology as Atlas AIDR, a new module inside its Atlas platform. The framing from CEO James Foster was refreshingly blunt: "Every enterprise is racing to put AI to work, and almost none of them can tell you what their AI agents are doing with their proprietary data, or what they cost."

We agree with the premise. And when a top-tier MDR player acquires a company for the express purpose of answering "what AI do we actually have, and what is it spending?", a problem that lived on the edge of the roadmap has moved to the center of the market. This is the shadow AI problem - and it's the same discovery gap we've been mapping for clients since before it had a tidy name.

What eSentire actually shipped

Stripped of the launch language, Atlas AIDR does four things that matter to a CIO and a CFO:

• A live AI inventory. Not a spreadsheet from a one-time survey, but a running record of the agents, models, providers, MCP servers, and tools in use - tied back to specific users, teams, and workstations.
• Inline controls. Redacting PII before it reaches a model, screening for prompt injection, and enforcing policy rules at the point of use rather than after the fact.
• Agent supply-chain governance. Watching the external MCP servers and tools that autonomous agents call, including when a server's description or behavior quietly changes.
• Cost and usage accounting. Token economics per request, rolled up by user, team, and model, with budgets, alerts, and anomaly digests.

Notably, eSentire built this to be vendor-agnostic and stressed there's no rip-and-replace - it's designed to sit alongside the AI providers and security tools customers already run. Hold that detail; it matters more than the acquisition itself.

The market context is the part nobody disputes. Citing McKinsey, eSentire notes that 76% of employees now use AI at work, up from 30% two years ago. Adoption has outrun governance, and the gap is widening.

Why this is a signal, not just a product launch

Discovery used to be the unglamorous first step nobody wanted to fund. It's now table stakes for an MDR platform. That shift tells you where buyer urgency is heading.

For anyone who has watched shadow IT play out over the last decade, the pattern is familiar. First the problem is invisible. Then it's a niche concern owned by a few worried architects. Then a major platform vendor puts a name and a SKU on it - and suddenly it's a line item every CIO and CFO is expected to have an answer for. Atlas AIDR is that moment for shadow AI.

Shadow AI is shadow IT, accelerated

Here's the number we keep coming back to from our own engagements: when we run a discovery pass, 30-40% of the SaaS applications in a client's environment are ones they didn't know they had. That was true before generative AI. It is worse now, and for a structural reason.

Every SaaS app your teams already use is bolting on AI features. On top of that, employees are spinning up standalone agents, connecting browser-based assistants, and wiring MCP connections to internal and external tools - often with a corporate credit card and no procurement ticket. The unknown surface isn't just growing; it's compounding, because each agent can call other tools and providers you never evaluated.

You cannot govern, secure, or budget for what you can't see. That's the whole ballgame.

The CIO problem: no inventory, no control

For the CIO, this is a visibility and sprawl problem before it's a security problem. Which agents touch regulated data? Which providers is that data leaving to? Which MCP servers do your agents trust - and who approved them? Without a continuously updated inventory, every AI policy is aspirational. You end up with a governance document that "looks good in a board report," to borrow a phrase from eSentire partner Nim Nadarajah of CriticalMatrix, without a record of what's actually happening in the business.

The CFO problem: unmanaged, variable AI spend

For the CFO, shadow AI is a spend-control problem hiding inside other people's invoices. Token-based and consumption pricing means your AI bill is variable by design, it scales with usage you can't currently attribute, and it's frequently buried inside existing vendor charges rather than sitting on its own line. If you can't tie spend to a team, an owner, and an outcome, you can't set a budget - and you certainly can't catch the runaway agent before it shows up on the statement. The fact that eSentire built cost accounting, budgets, and anomaly alerts directly into a security module is the clearest tell that AI spend has become a finance-office concern, not just an IT one.

Where a vendor-neutral view differs from a platform module

Atlas AIDR is a genuinely useful capability - if you're standardized on eSentire's platform. But most enterprises aren't standardized on anything. They run multi-vendor estates: several security tools, multiple AI providers, a dozen SaaS suites each shipping their own AI, and agents that cross all of it. A module that sees the world through one platform's telemetry gives you one platform's view.

This is where our position differs on purpose. Cortexa doesn't sell you the security stack and then also grade its own homework on what that stack can see. As a vendor-neutral advisor, our job is to build you an independent, complete inventory across every AI provider, SaaS app, and agent - regardless of which MDR, SIEM, or endpoint vendor you've chosen - and to turn that inventory into decisions about risk, sprawl, and spend.

Read the eSentire news the right way: not as "which platform wins," but as confirmation that you need an answer to what AI you have and what it costs. The follow-up question is whether that answer should be owned by a vendor with something to sell you, or held independently.

What to do this quarter

You don't need to wait for a platform migration to act. Start here:

1. Stand up a live AI inventory - agents, models, providers, MCP servers, and tools - and treat it as continuous, not a one-time survey.
2. Attribute usage and cost to teams and named owners so both the CIO and CFO are reading from the same record.
3. Map data flows. Know which proprietary and regulated data is going to which models and providers.
4. Put budgets and anomaly alerts on AI spend before the invoice does it for you.
5. Create an approval and offboarding path for new agents and MCP connections, so the inventory stays clean.
6. Pressure-test your visibility. Is it tied to a single vendor's platform, or is it independent enough to survive a change in tooling?

The takeaway

The best line from the eSentire announcement came from its partner, not its press release: "You can delegate work to AI. You can't delegate accountability." That's exactly right. The market now agrees that shadow AI is real, expensive, and ungoverned. The open question for every CIO and CFO is whether they'll get their inventory and their cost controls from a single vendor's platform - or from an independent view that spans the whole estate.

If you want to know what your AI agents are actually doing with your data and what they're costing you - without buying another platform to find out - that's the work Cortexa does. Let's start with a discovery pass and see what turns up. In our experience, it's usually more than you expect.