Head-to-head · updated 13 September 2026

DataShield vs Styra / OPA: who runs the policy engine now that Styra has gone quiet?

Open Policy Agent is the best general-purpose policy engine in open source, and we're not going to pretend otherwise. It graduated from the CNCF in 2021, it runs Kubernetes admission control at Goldman Sachs and Capital One, and Rego can express policy that most vendor rule builders can't. If you already run OPA, keep it.

But OPA answers one question: is this allowed? It doesn't issue the agent's identity, hold its credentials, pull its authority mid-job, or keep proof that the decision happened. You build that. DataShield ships it. And there's a new wrinkle: Styra, the company that sold the commercial layer, has effectively wound down. Details below, with sources.

Get free Demo Center access Get an instant quote

The short version

Pick DataShield when

You need the whole agent seam, not just the decision. SSO and passkeys, MCP tool tokens with scope ceilings, a Connection Vault for credentials, and a signed audit chain. See how Auth fits together.
An auditor or the EU AI Act's Article 12 will ask you to prove the log wasn't edited. Our chain answers with math. OPA's decision logs are plain JSON shipped to a service you choose. Run the verifier.
You want to pull an agent's authority mid-session and have the next governed tool call fail, not wait for a token to expire.
You'd rather buy the assembly than staff it. A team that wires OPA to their own identity, vault, and evidence layer is signing up for a year of platform work.

Pick Styra / OPA when

Your policy problem is Kubernetes admission control, Terraform gates, CI/CD checks, or Envoy authorization. That's OPA's home turf and we don't play there at all.
You need policy depth. Rego handles partial evaluation, hierarchical data, and rules we simply can't express in Cedar. Ten years and 450-plus contributors show.
Free and Apache-2.0 matters more than support. OPA costs nothing, embeds anywhere, and CNCF graduation means no single vendor can change the terms on you.
You have a platform team who wants to build the surrounding plumbing, and enough of them to keep it running in three years.

Bottom line: OPA decides. DataShield decides, issues, revokes, and proves. Run both if you like: keep OPA for your clusters, put DataShield on the agent path where the evidence obligation lands.

Feature by feature

Competitor cells describe what the OPA project's public site, docs, and repos say as of the date above. If we've mischaracterised something, email support@myorg.ai and we'll correct it, credited.

What matters	DataShield	Styra / OPA	Edge
Audit evidence	SHA-256 hash chain with Ed25519-signed checkpoints that are themselves chained. Verification names the failure: tampering, insertion, deletion, or truncation. Public verifier.	Decision logs as JSON, shipped to an HTTP service, with a Rego masking policy for sensitive fields. The docs describe no cryptographic tamper evidence. Whatever you ship them to owns that problem.	◆
Agent authorization	Every governed tool call passes a scope ceiling, a declared authority tier, and a revocation re-check before dispatch. Cedar covers admin, platform, config, and token decisions.	A decision engine that answers whatever question you send it. There is no agent identity model, no tool token, and no revocation concept. You supply all three.	◆
Policy language	Cedar, with policy_explain and entitlement expansion. Deliberately narrow. Easier to read, harder to get clever with.	Rego. Partial evaluation, hierarchical data, rule indexing, a linter (Regal), a playground, and editor plugins. Version 1.20.2 shipped 3 September 2026 with new `and` and `or` keywords. Genuinely better as a language.	◇
Break-glass	Scoped, time-boxed emergency access for agents that auto-revokes and can't be quietly dropped from the log.	Not a concept in the engine. You'd write a Rego rule and build the workflow, expiry, and log around it.	◆
GDPR erasure	Crypto-shred of per-subject key material plus ISO 27560 consent receipts. Actor IDs are HMAC-committed, so the chain still verifies after erasure.	Out of scope. A policy engine holds no subject data to erase, which is fair, but it also means the erasure story lands entirely on you.	—
Data handling and tokenization	Datasets are tokenized at ingest; agents query tokenized data over MCP; detokenization is a privileged, audited operation. Plus quasi-identifier generalization (dates to year, decade or age band; ZIPs to 3 or 4 digits; partial phones, SSNs and emails) with a measured cardinality-reduction score per column.	None. OPA evaluates data you hand it. Enterprise OPA added data filtering for database-backed workloads, and that code is now donated to the community.	◆
MCP and agents	Native MCP endpoints on Auth, Ontology, and Lighthouse. MCP tool tokens with scope ceilings, and per-call metering attributed to the agent.	One MCP entry in the ecosystem: a third-party server for authoring and debugging Rego, filed under tooling, layer editor. It helps an LLM write policy. It doesn't authorize an agent.	◆
Kubernetes and infrastructure policy	Nothing. We don't do admission control, Terraform gates, or CI/CD policy, and we won't pretend to.	The default choice. Gatekeeper, Conftest, Envoy, Terraform, Docker, Kafka, SSH and sudo. A decade of integrations.	◇
Deployment	Self-hosted in your own cloud or data center, or on a dedicated single-tenant server we operate. Your keys.	Anywhere. Sidecar, Go library, standalone service, or WASM. Bundles can be digitally signed and OPA refuses to activate one that fails verification, which is a nice touch.	◈
Vendor and support	Small team, and we say so. Design-partner terms include source escrow so a small vendor isn't a single point of failure.	The OPA creators and much of the Styra team joined Apple in August 2025. Enterprise OPA, the Control Plane, the SDKs, and Regal were donated to the CNCF org. Styra DAS gets no mention in that post. styra.com no longer resolves in DNS. The project's own support page lists two third-party firms, and says the list isn't vetted.	◆
Maturity signals	Auth, Guardian and Lighthouse are live in production (Guardian and Lighthouse since April 2026). SOC 2 not yet certified, and we say so.	CNCF graduated January 2021. 12.2k GitHub stars, monthly releases, named adopters including Goldman Sachs, Capital One, BNY Mellon, and Atlassian. The project is in far better shape than we are.	◇
Pricing	Published model, scoped instant quote, no sales wall.	OPA is free under Apache 2.0. The commercial tiers that used to sit above it have no published price and, as of this date, no reachable website.	◇

◆ DataShield leads◇ Styra / OPA leads◈ comparable

OPA claims are drawn from openpolicyagent.org, the CNCF project page, and the open-policy-agent GitHub org, last checked 13 September 2026. The Styra DNS check was run against 1.1.1.1 and 8.8.8.8 on the same date.

Three things you get here that you won't get from a policy engine

An identity for the agent, not just a decision about it

A policy engine takes a JSON input and returns allow or deny. Something upstream has to decide who the caller is, mint its token, cap its scope, and hold its database credentials. That's Auth: SAML, OIDC, passkeys, RFC 9068 tokens, JWKS rotation, and a Connection Vault. See the pieces.

Proof the decision log wasn't edited

OPA's decision logs are honest JSON. They're also as trustworthy as the store you ship them to. Ours is a hash chain with signed checkpoints, and the verifier tells you what broke, not just that something did. That's the property EU AI Act Article 12 and HIPAA §164.312(b) reviewers push on. Try it in your browser.

Authority that changes mid-flight

An analyst quits on a Tuesday. Their agent is 20 minutes into a 40-minute job. With DataShield the next governed tool call is re-checked against current authority and fails closed. With a bundle-distributed policy, the fix ships on the next bundle poll. How the architecture handles it.

Where Styra / OPA is genuinely stronger

Rego is a better policy language than Cedar. That isn't close. Partial evaluation, hierarchical data, rule indexing, and a linter written partly in Rego itself are the fruit of ten years and more than 450 contributors. The integration surface is enormous: Gatekeeper, Conftest, Envoy, Terraform, Kafka, SSH. CNCF graduation gives it neutral governance that no vendor-controlled engine has, Cedar included, and Cedar is the one we ship. If your job is cluster policy, use OPA and stop reading.

Here's the push-back. Between the engine and a governed agent sits a lot of work nobody hands you: identity issuance, token scoping, revocation that lands in seconds, credential storage, break-glass, and evidence an examiner will accept. Teams reach for Styra DAS to close that gap. As of today styra.com doesn't resolve, the commercial code has been donated to the community, and the Enterprise OPA README asks whether anyone wants to maintain it. The engine is thriving. The product that wrapped it is not, and a platform lead should know that before the renewal conversation they were expecting.

Questions worth asking both of us

These are the questions we'd want answered if we were buying. Ask them on every vendor call, including ours.

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, and verification tells deletion apart from truncation and tampering. Run it at /verify. OPA: its decision log docs cover masking and shipping events to an HTTP service, with no tamper-evidence mechanism described. The integrity question moves to whatever sink you pick, so ask that vendor instead.

What happens to a revoked agent mid-session?

DataShield re-checks authority on every governed tool call, so revocation bites on the next call. OPA has no agent or session model. Revocation means changing input data or pushing a new bundle, and how fast that lands depends on your bundle poll interval and your own plumbing. Time that path before you trust it.

Is Styra still in business, and what happened to Styra DAS?

On 20 August 2025 the OPA creators announced that they, and many Styra team members, had joined Apple. The same post moved Enterprise OPA, the new OPA Control Plane, the SDKs, and Regal into the CNCF OPA GitHub org. Styra DAS isn't mentioned anywhere in it. As of 13 September 2026 styra.com returns SERVFAIL from public DNS resolvers, so we can't check their side of the story. If you're a DAS customer, that's a question for your account team, assuming you can still find one.

So is OPA itself at risk?

No, and we'd rather you didn't read it that way. OPA is CNCF graduated with no governance or licence change, the maintainer list is intact with employers updated to Apple, and releases have kept a monthly cadence through v1.20.2 on 3 September 2026. Commercial support is now third-party: the project's support page lists two firms and notes it isn't vetted by maintainers. That's a procurement fact, not a health warning.

Why Cedar instead of Rego?

Because we wanted policies a compliance reviewer could read without learning a new language, and an engine with decidable analysis. Rego wins on expressive power. We'd rather lose that fight than ship policy nobody can review. Note what we don't claim: Cedar fires on admin, platform, config, and token decisions in Auth, not on every tool dispatch. Tool calls go through scope ceilings, authority tiers, and a revocation re-check. Anyone telling you their policy language evaluates every agent action should be asked to show it live.

Does DataShield have SOC 2?

Not yet, and we won't imply otherwise. Auth is live a public threat model and a verifier anyone can run. Guardian and Lighthouse have been 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.

"the creators of Open Policy Agent (along with many team members from Styra) have joined Apple." Enterprise OPA, OPA Control Plane, the SDKs, and Regal move to the CNCF OPA GitHub org. Styra DAS is not mentioned. — Open Policy Agent blog, 20 Aug 2025
"EOPA has been donated to the OPA community… If you are interested in maintaining this project, please reach out to the maintainers." — open-policy-agent/eopa README, 13 Sep 2026
OPA reached CNCF Graduated maturity on 29 January 2021. — CNCF project page, 13 Sep 2026
Commercial support for OPA is listed as two third-party firms, with the note that the list "is not vetted by the OPA project maintainers." Styra is not listed. — openpolicyagent.org/support, 13 Sep 2026
Decision logs support Rego-based masking of sensitive fields and delivery to an HTTP service. No cryptographic tamper evidence is described. — OPA docs, decision logs, 13 Sep 2026
OPA v1.20.2 released 3 September 2026; monthly release cadence intact. Repo shows 12.2k stars. — GitHub, open-policy-agent/opa, 13 Sep 2026

Other head-to-heads

Policy

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 →