Your AI Agents Can Reach Data Nobody Classified. Who Decides What's Sensitive?
A $150M bet says agent access starts with knowing your data
Most organizations treat AI agent access as an identity problem. Give each agent the right role, scope its credentials, and it will only touch what it should.
The larger problem is that nobody knows what most of that data is. On November 4, Sentra announced a $150 million investment led by Capital One Ventures to control what AI agents can reach. Its CEO, Yoav Regev, put the issue plainly: "AI agents are now the busiest consumers of enterprise data, and many of them inherit access to data that hasn't been fully classified or governed."
That puts a recurring job on your data team's desk. Every time an agent is connected to a new source, someone has to answer which tables, files and fields inside it are sensitive, who owns them, and where they flow next.
And it gets harder as you scale. One agent reading one CRM is a review. Dozens of agents across shared drives, warehouses and SaaS tools mean thousands of assets nobody has labeled, each one reachable the moment an agent switches on.
As agents become the main users of enterprise data, every agent needs to know what it can reach, who owns it and which other agents it can trust. That is what we are building at OpenMap: The Agent Network, the connective tissue for the world's AI agents, starting with every person and every local business. We are raising a £5M seed. Meet us at openmap.ai/investors.
Why Can't Anyone Say What an Agent Is Allowed to See?
Picture a finance operations agent given read access to a shared analytics workspace so it can reconcile invoices.
The workspace also holds a payroll extract someone copied there last spring, a customer table with unmasked account numbers, and a contract folder nobody has opened in two years.
The agent does nothing wrong. It simply has access to all of it.
This is not an access control problem. It is a data knowledge problem.
The Cost Isn't One Breach. It's Every Approval That Stalls.
Lyft's CISO, Chaim Sanders, described the position most teams are in, in the same announcement: "we need to know what data our agents can reach before we turn them on."
That is the right instinct. It also turns every new agent into an investigation.
Before a single agent can be approved for a data source, someone usually has to:
Find every dataset, file and dashboard the agent's credentials can reach
Work out which of those assets hold personal, financial or regulated data
Identify who owns each one and who can approve agent access
Trace where the data is copied, joined or exported downstream
Decide what happens to the copies nobody remembers creating
None of these tasks are particularly difficult. They are slow, they cross team lines, and they usually land on the few engineers who already know where everything lives.
So agent rollouts wait. Or they go ahead with broad access and the hope that nothing sensitive was in reach.
How to Know What Your Agents Can Reach in 30 Days?
You do not need to redesign your data platform, or replace your security stack, to know what your agents can reach.
Within the first month, focus on the sources your agents already touch or are about to: the warehouse, the CRM and the shared drives behind your first copilots.
Start by building an inventory of what lives in each source, and map how the most sensitive datasets move into reports, extracts and other systems. Then name one owner per critical dataset, with the authority to approve agent access.
Next, expose this context through the catalog or policy layer your security and identity tools already read, so access decisions are made against known data, not guesses.
DataManagement.AI accelerates this process by automatically discovering metadata, mapping lineage across the modern data stack, identifying downstream dependencies, and creating a centralized knowledge layer that stays continuously updated as your environment evolves. Ownership and policy travel with every entity, so when an agent requests access, the owner can see what that data is and where it goes next.
Instead of spending months documenting assets manually before the first agent goes live, your data and security teams start from the same map. Agent approvals move at the speed of a review, not an excavation.
Agent Access Risk Is Actually a Data Inventory Problem
Identity controls, agent monitoring and data security platforms will keep improving. Sentra's raise is one sign of how much money is now chasing the question of what agents can reach.
But every one of those controls depends on an answer it cannot invent for itself: what is this data, and who is responsible for it?
Organizations that scale agents safely do not just tighten permissions. They know what sits behind every permission they grant.
Warm regards,
Shen and Team