Part 1 split AI agents into two kinds and looked at the controls BigID and Varonis are building for them. Part 2 covered what protecting the data does for each kind, and where it stops. This part is about how to tell, during an evaluation, which kind of tool you’re actually looking at.
DLP, DSPM, and data-centric protection do different jobs
A lot of tools get called data security, and buyers often assume they overlap more than they do.
Data loss prevention (DLP) watches the ways data leaves your company, like email, web uploads, and USB drives, and blocks or flags what looks sensitive. Data security posture management (DSPM) scans where data is stored, classifies it, and flags exposure, such as a folder anyone in the company can open. Both find data and watch what happens to it, and that’s valuable.
Data-centric protection acts on the data once it’s found. It encrypts, masks, redacts, quarantines, or deletes it, so the file is protected without waiting on someone to respond to an alert.
One doesn’t replace the other. Going by their own published materials, neither BigID nor Varonis encrypts data natively. BigID orchestrates encryption through separate tools, and Varonis’s core platform discovers, classifies, and monitors without native encryption. Both do a good job of showing where sensitive data lives and who or what touched it, and both do things PK Protect doesn’t, including DLP and governing agent access. Neither is designed to make a file unreadable to an agent that reached it without permission.
Five questions to ask in any DSPM or DLP evaluation
When the tool flags that someone or something reached a sensitive file it shouldn’t have, what happens to the file? If the answer is that an alert fires, ask what that person or agent can do with the contents before someone acts on the alert.
Is encryption built into the platform, or does it come from a separate tool? A separate tool is one more thing to buy, deploy, and keep in sync, and any gap between discovery and protection leaves classified data unprotected.
Does the protection stay with the file when the file moves? Files get copied and exported to places nobody planned for, and they’re only protected there if the protection is part of the file.
Does it cover mainframe and on-prem systems, or only cloud and SaaS? Many DSPM tools started in the cloud. If your sensitive data is on z/OS, IBM i, or file servers in your own data center, ask about those specifically.
How does it handle each kind of AI agent as their numbers grow? For assistants working on a person’s behalf, ask how it helps you cut down what each person can reach, since the assistant can reach the same things. For agents with their own identity, ask whether each new one has to be found and given permissions before it’s covered, or whether the data stays unreadable to it either way.
Why the first question matters
A readable file in the wrong hands is expensive. IBM’s 2025 Cost of a Data Breach Report put the average U.S. breach at $10.22 million. The number of identities that could reach that file is growing too. Machine identities now outnumber human ones about 109 to 1, and 40% of them can reach company data.
A DSPM or DLP tool can flag a file and still leave it readable to whatever copies it. Changing that takes something that protects the data itself.
Where PK Protect fits
PK Protect is built for that part of the problem. It finds sensitive data on endpoints, in the cloud, in Microsoft 365, and on z/OS and IBM i, and classifies it under one classification policy. From there, your team can encrypt, mask, redact, quarantine, or delete what it finds. Protection stays on the file after it moves, so an agent running under its own identity can’t read an encrypted file without access to the key. For assistants working on a person’s behalf, cleaning up overexposed data and masking or redacting values people don’t need limits what those assistants can surface. Format-preserving encryption keeps protected data usable for the systems that depend on it.
To be clear about the limits, PK Protect won’t stop a file transfer while it’s happening, it won’t tell you what your agents are doing, and it won’t stop an assistant from reading what its user is allowed to read.
If you’d like to see what sensitive data is in your environment today and what it would take to protect it, request an assessment.
Sources
PKWARE vs. BigID and PKWARE vs. Varonis, PKWARE comparison pages drawn from each company’s published materials.
Cost of a Data Breach Report 2025, IBM.
2026 Identity Security Landscape, Palo Alto Networks.