Head-to-head · updated 13 September 2026

DataShield vs Microsoft Purview: who proves what the agent did?

Microsoft Purview is very good at the thing it's built for. Inside an Azure and Microsoft 365 tenant, it maps the estate. It labels the sensitive bits. It runs DLP. It gives the compliance team Audit, eDiscovery, and Records. All in one portal. Unified Catalog adds governance domains and data products. Also a live glossary, quality scores, and OKRs. Say your data lives in SharePoint and OneLake. Say your agents are Copilots. Purview is the default answer. It's a fair one. We won't pretend otherwise.

Ours is a smaller, narrower thing. We run where you put us. We tokenize datasets at ingest. Agents query the tokenized data over MCP. Reversing a token is a special, audited step. We recheck each checked tool call. We check it against the agent's current access, before it runs. Every choice seals into a hash chain. You can check that chain, without trusting us or Microsoft. Below is the honest split. It includes the many rows Purview wins.

Get free Demo Center access Get an instant quote

The short version

Pick DataShield when

Someone will ask you to prove an agent's log was not edited. Not search it. Prove it. Our chain answers with math. Run the verifier.
You need to pull an agent's access mid-session. You need the very next tool call to fail closed.
The agents that matter are not Microsoft agents. Purview's own matrix marks Claude Enterprise and ChatGPT Enterprise agents as not supported. That's true for both info protection and compliance.
The policy engine, the vault, and the proof have to run on your own system, on keys you hold.

Pick Microsoft Purview when

Your data and your agents both live in the Microsoft tenant. Nothing third-party matches that reach into SharePoint, Exchange, Teams and Copilot.
The compliance office is buying. Audit, eDiscovery, Records Management, Communication Compliance and Compliance Manager are real products, with real users.
You want a business-facing catalog. Think governance domains, data products, critical data bits, quality scores, OKRs and health checks. Ours is thinner. It only covers PostgreSQL.
It's already on the bill. E5 money is spent money. That argument wins a lot of meetings.

Bottom line: Purview governs the Microsoft estate. We govern the data an agent touches and keep proof of every decision, wherever that agent runs. Most of our buyers keep both, and the line between them is the tenant boundary.

Feature by feature

Competitor cells describe what Microsoft's public docs say. That's as of the date above. If we got something wrong, email support@myorg.ai. We'll fix it, and credit you.

What matters	DataShield	Microsoft Purview	Edge
Audit evidence	A SHA-256 hash chain, with Ed25519-signed checkpoints. The checkpoints chain too. A check names the failure: tampering, insertion, deletion, or truncation. Try the verifier.	The unified audit log: searchable, exportable, and kept 180 days by default. Premium holds Entra, Exchange, OneDrive, and SharePoint records a year. Ten years needs a per-user add-on. It doesn't apply to old records. We found no crypto tamper proof.	◆
Agent authorization	Every checked tool call passes three gates. A scope ceiling. An access tier. A check for revoke. The call fails closed.	Labels, DLP rules, and audit capture around named apps. Per their docs, agents inherit the protections of their parent AI app. Nothing more. We found no per-call decision point for an agent.	◆
Break-glass	Scoped, time-boxed emergency access for agents. It auto-revokes. No one can quietly delete it from the log.	Privileged Access Management exists for admin tasks. Nothing agent-shaped that we could find.	◆
GDPR erasure	We crypto-shred each subject's keys. We add an ISO 27560 consent receipt. The audit chain still checks out after.	Retention rules, Data Lifecycle Management, and eDiscovery-driven deletion. That's a workflow. It isn't a crypto check.	◆
Tokenization	Set, join-safe tokens, made at ingest. Held in a vault we can reverse. Plus quasi-identifier generalization. Dates cut to year, decade, or age band. ZIP codes cut to 3 or 4 digits. Phones, SSNs, and emails cut too. Each column gets a measured score for the cut. Reversing a token is a special, logged step.	Sensitivity labels and DLP decide who may read the value. The raw value stays where it is. Purview doesn't pseudonymize and reverse.	◆
MCP and agents	More than 200 MCP tools across Ontology, Auth, Corpus and Lighthouse. Auth issues MCP tool tokens with scope ceilings. It meters each call against the agent.	No mention of MCP server support. No MCP tool-call audit either. Not in the Purview docs we read. Agent checks run through Entra registration and the Copilot surfaces.	◆
Non-Microsoft agents	Any MCP client. Any IdP we federate to over SAML or OIDC. Nothing about the control plane assumes a vendor.	Their own agent matrix marks Anthropic Claude (Enterprise) agents not supported. That's true for both columns. ChatGPT Enterprise agents aren't supported either. Except for classification and Insider Risk.	◆
Catalog and discovery	We scan, profile, and classify a live PostgreSQL source in place. No rows leave it. That's against 129 field classes. Think PII, PHI, money data, and secrets. We build a glossary, not just link to one. We add typed lineage with per-hop access checks. Plus a 29-command steward workflow. PostgreSQL today, not your whole estate.	Data Map scans Azure sources. Plus S3, Redshift, BigQuery, Snowflake, Oracle, SAP, and more. Unified Catalog adds governance domains and data products. Also critical data bits, quality rules, and OKRs. Far broader than us.	◇
Microsoft 365 reach	None. We don't see SharePoint, Exchange, Teams or Copilot prompts.	Deep. No one else can match it. Prompts and answers land in the unified audit log. They carry the labels of the files touched.	◇
Compliance suite	Consent receipts, erase steps, retention, and a signed proof ledger. No eDiscovery. No records management. No communication compliance.	Audit, eDiscovery, Records Management, Data Lifecycle Management, Communication Compliance and Compliance Manager. All of it. Plus templates for AI rules.	◇
Deployment	Self-hosted, in your own cloud or data center. Or a dedicated single-tenant server we run. Docker images for Auth, Ontology, Corpus, and Lighthouse. Ed25519 audit-signing keys stay on your instance. They sit in a key folder you control. An unknown key fails closed. KMS or HSM custody is a built seam in Auth. It's not a shipped link. HMAC tokenization keys sit in your setup today. Not in a KMS.	SaaS, inside your Azure tenant. Non-Azure sources get scanned through a self-hosted runtime you install. But the control plane stays in Azure.	◆
Pricing	Published model, scoped instant quote, no sales wall.	Metered by use. Unified Catalog charges per checked asset per month. A processing unit buys 60 minutes of compute. Audit charges per 1,000 records ingested, plus more. The public page blanks the dollar figures. You must sign into the calculator to see them.	◆
Maturity signals	Auth, Guardian and Lighthouse are live in production. Guardian and Lighthouse have run since April 2026. SOC 2 is not yet certified. We say so.	Microsoft. Every cert you can name, at platform level. A reference list longer than our roadmap.	◇

◆ DataShield leads◇ Microsoft Purview leads◈ comparable

Purview claims come from learn.microsoft.com. Also the Azure pricing page. We last checked 13 September 2026. We link them below, rather than work from memory.

Three things you get here that you won't get from a platform catalog

Evidence, not a search box

The unified audit log is a good search box. It isn't proof. An examiner asks: were the March records edited? You get an export, and your word for it. Ours is a hash chain with signed checkpoints. The verifier says what broke. Not just that something did. That's the trait EU AI Act Article 12 and HIPAA §164.312(b) reviewers care about. Try it in your browser, no signup.

Authority that can change mid-flight

An analyst leaves on a Friday. Their agent is 20 minutes into a 40-minute job. With DataShield, the next checked tool call gets rechecked. It's checked against current access. It fails closed. A token lifetime, by design, doesn't do that. How Auth does it.

The value is gone, not gated

A label decides who may read the name. Tokenizing at ingest means the name isn't there to read. The token still joins across datasets. So your analysts keep working. The raw value comes back only through a vault call. That call is special, and audited. We wrote up the mechanism split in more detail here.

Where Microsoft Purview is genuinely stronger

Start with the obvious. Purview sees things we never will. Prompts and answers from Microsoft 365 Copilot land in one place: the unified audit log. They carry the sensitivity labels of the files touched. No third party gets that. Not without the same Graph and Entra access Microsoft has. Unified Catalog is a real catalog for real business users. It has governance domains, data products, and critical data bits. Also glossary terms that carry policy. Plus quality scores, OKRs, and health checks. Our catalog covers PostgreSQL and nothing else. Their Data Map reaches S3, Redshift, BigQuery, Snowflake, Oracle, and SAP. And the compliance suite around it is a set of products. People have run them for years. Think Audit, eDiscovery, Records. One portal. One bill. One vendor.

Here's the push-back. It's narrow on purpose. Two of Microsoft's own pages set the line. The agent matrix marks Anthropic Claude (Enterprise) agents as not supported. That's true for info protection. It's true for compliance too. It marks ChatGPT Enterprise agents as supported. But only for two things: classification and Insider Risk. So the story covers Microsoft's own agents well. It covers other agents barely. Meanwhile the retention clock runs. 180 days by default. A year for a few workloads on Premium. Ten years, only with a per-user add-on. That add-on skips old records. Records from service principals and system events are fixed at one year. That's what an agent makes, not a person. The doc says you don't set that period. If you need proof past that point, export it. Keep it yourself. Microsoft's own storage guidance puts it plainly. You must store those logs long-term, yourself. That do-it-yourself step is our product.

Questions worth asking both of us

These are the questions we'd want answered, if we were the buyer. Ask them on each call. Ours included.

Can you cryptographically prove an audit log entry wasn't deleted?

DataShield: yes. Each record commits to the one before it. Checkpoints are signed and chained. A check tells deletion apart from truncation and from tampering. Run it against a sample chain at /verify. Purview: the docs describe search, and export to CSV. Also the Management Activity API. We found no tamper-proof mechanism. Ask them to show one. Then ask why the export path caps at 50,000 records per search, on Standard.

What happens to a revoked agent mid-session?

DataShield rechecks access on each checked tool call. So a revoke lands on the next call. Purview isn't the enforcement point here. Entra is. Entra works on token lifetime, and a Conditional Access recheck. Ask how long a compromised agent keeps working. That's after you pull its access.

How does GDPR erasure interact with the audit trail?

DataShield crypto-shreds each subject's keys. It issues an ISO 27560 consent receipt. Actor names in the chain are HMAC-committed. So the proof still checks out once the subject is gone. Purview offers retention rules, Data Lifecycle Management, and eDiscovery deletion. Those are workflows over a copy of the data. Ask for the mechanism, not the workflow.

Does Purview support MCP, and does DataShield replace it?

No, and no. We read the Purview docs for AI and agents. We found no MCP server. No MCP tool-call audit. No protocol-level agent checks. Governance flows from Entra registration and the Copilot surfaces. We ship more than 200 MCP tools. They span Ontology, Auth, Corpus, and Lighthouse. Our tool tokens carry a scope ceiling. That doesn't make us a replacement. Keep Purview for the Microsoft estate. Run us for the datasets agents query. And for the agents Purview marks as not supported.

We already pay for Purview in E5. Why add anything?

Often, you shouldn't. Say your agents are Copilots. Say your data is in the tenant. Spend the E5 money and stop. Three cases bring people to us anyway. An agent stack that isn't Microsoft. A retention rule longer than the audit clock. Or a regulator who wants proof, not a search result. One more thing. The Purview meters sit apart from the seat price. The public pricing page blanks the meter figures. You must sign in to see them.

Does DataShield have SOC 2?

Not yet. We won't say otherwise. Auth is live, with a public threat model. Anyone can run our verifier. Guardian and Lighthouse have run in production since April 2026. Design-partner terms include source escrow. So a small vendor isn't a single point of failure. Details on the security page.

Purview's own category line: "Unified data security, governance, and compliance solutions for the era of AI." Data Map and Unified Catalog are the two governance products. - Microsoft Learn, doc dated 15 May 2026
Agent control is inherited, not agent-specific: "AI agents typically have the same security and compliance protections for AI interactions as their parent AI app." Anthropic Claude (Enterprise) agents are marked as not supported, for both info protection and compliance. And ChatGPT Enterprise agents support only data classification and Insider Risk. - Microsoft Learn, doc dated 27 May 2026
Audit retention: 180 days by default. Audit Premium keeps Entra, Exchange, OneDrive, and SharePoint records one year. Ten years needs a per-user add-on licence. It doesn't apply to old records. Non-user records are fixed at one year. Think service principal and app activity. "This retention period isn't configurable." - Microsoft Learn, doc dated 18 May 2026
Audit export caps: "you can export up to 50,000 records to a .csv file from a single audit search query." For Audit (Premium), the export limit rises to 1,000,000 records. Over the cap, the file "might omit some audit logs." - Microsoft Learn, doc dated 19 June 2026
Microsoft's own storage guidance on keeping log evidence: "You are responsible for storing those logs persistently, as might be required for regulatory or other purposes." - Microsoft Learn, Azure immutable blob storage, 18 Aug 2026
Purview pricing is metered by use. Unified Catalog charges per checked asset per month. A Data Governance Processing Unit buys 60 minutes of managed compute. Audit charges per 1,000 records ingested. The public page renders the dollar figures as blanks. Sign into the Azure calculator to see them. - Azure pricing, checked 13 Sep 2026

Other head-to-heads

Same market

Get free Demo Center access

You've seen the proof

Ready for a number? Scope your deployment and we'll price it against your own economics.

Get your quote →