Privacera (Trust3 AI)

Overview

Privacera is data access governance built by the people who created Apache Ranger, and that lineage explains both its architecture and its breadth.

Know the naming before you search: it rebranded to Trust3 AI in March 2026, with much of privacera.com now redirecting to trust3.ai, while the product documentation stays under the Privacera name.

This is a rebrand rather than an acquisition. The legal entity is unchanged, trading as Trust3 AI, which is worth stating plainly because the change is easily misread.

Its real differentiator against Immuta is that it offers three distinct enforcement engines and is open about which applies where, rather than presenting one mechanism.

Ranger plugins enforce in-engine in real time for open-source engines and Databricks clusters. PolicySync pushes policy into native primitives on Snowflake, Redshift, Lake Formation and Unity Catalog.

That distinction matters operationally: PolicySync is a polled push on an interval, not an inline decision at query time, so policy changes take effect with some latency.

Object storage is handled by DataServer, which by default issues signed URLs rather than sitting in the path, though an optional proxy mode does put it inline.

Policy supports role, attribute, tag and purpose-based models, with tag-based control the reuse mechanism: change a classification and enforcement follows automatically everywhere that tag appears.

Connector breadth is its other advantage, with 28 documented connectors against Immuta's eight, spanning warehouses, Spark, query engines and all three major object stores, though marketing claims more than the docs list.

Watch two practical details: audit retention defaults to two weeks on the cloud service unless you export it, and no funding round has been published since March 2021 despite the rebrand and repositioning.

Key features and capabilities

The same headings are used for every access policy enforcement entry, so two tools can be read side by side.

How policies are written

• Four access control models, role-based, attribute-based, tag-based and purpose-based
• Policies are authored centrally against the Apache Ranger policy model, with security zones and roles
• Tag-based control is the reuse mechanism, driving access from classification rather than resource names
• Tag changes propagate automatically, so reclassifying data updates enforcement everywhere it appears
• The rebranded platform adds natural-language policy creation and a conversational governance interface

How policies are enforced

• Three distinct engines rather than one, which is the clearest difference from its rivals
• Ranger plugins embed in the engine and enforce in real time, used for open-source engines and Databricks clusters
• PolicySync pulls policy on an interval then pushes it into native primitives, so it is not an inline decision
• PolicySync targets Snowflake, Redshift, Lake Formation and Unity Catalog, in native or view mode per connector
• DataServer covers object storage, issuing signed URLs by default with an optional inline proxy mode

Masking and de-identification

• Dynamic column masking and row filtering translated into each target's native constructs
• A separate encryption product with encryption and masking schemes, the latter irreversible
• Techniques named include hashing, tokenisation and value replacement
• Format-preserving encryption is published for anonymising data at rest
• Differential privacy and k-anonymisation are not published

Classification and discovery

• A first-party discovery service scanning structured and unstructured data to identify and tag it
• Classification covers personal, financial and other protected content, with automatic tagging
• Discovery and enforcement share one tag pipeline, so classification feeds policy directly
• A scan API allows programmatic discovery, with a compliance summary report
• Tags can also be consumed from Collibra, and the rebranded platform extends discovery to AI agents

Platform coverage

• 28 documented connectors, though marketing claims more than fifty
• Warehouses including Snowflake, Redshift, BigQuery, Synapse, Dremio, Vertica and Athena
• Lakehouse and Spark including Unity Catalog, Databricks SQL and clusters, EMR, Glue and Lake Formation
• Query engines Trino and Starburst, plus all three major object stores through DataServer
• Apache Iceberg support arrived in April 2026; there is no Fabric, Teradata or ClickHouse connector

Audit and monitoring

• A unified interface across platforms, with plugin audit logs sent to a central service
• Cloud and data-plane deployments retain audit for two weeks by default
• Self-managed retention is around ninety days, configurable on the audit server
• Documented export to your own S3 or Google Cloud Storage for long-term retention
• Anomaly detection on access is published only for the newer agent observability layer

Where it runs and what it costs

• Three options, fully managed cloud, self-managed in your own private cloud, or a hybrid data plane
• The hybrid is its own recommendation, keeping the interface hosted and data-touching components in your network
• Self-managed runs the entire platform inside your own private cloud
• Regions are not published, with documentation referring generically to any cloud provider
• No unit of pricing is published

Pricing

Price on requestQuote only; nothing published

Quote-only. No pricing page exists on either domain, and no tiers, currency, list prices or unit of metering appear anywhere across the marketing site, the rebranded site or the documentation. The published routes are demo requests on both. A self-service cloud trial does exist, with online registration, but its length is not published on any first-party page, so the thirty days often quoted elsewhere should not be relied on. No free tier is published.

Vendor pricing page →

Demos and videos

• Meet Trust3 AI, the next chapter for Privacera Product demo · YouTube
• Connector architecture explained Tutorial · YouTube

About Privacera, trading as Trust3 AI

Privacera was founded in 2016 by Balaji Ganesan, its chief executive, and Don Bosco Durai, its chief technology officer, both co-creators of Apache Ranger, with Neeraj Sabharwal. It rebranded to Trust3 AI in March 2026 under the same Delaware entity, now based in Newark, California rather than the Fremont address older sources give. It is private, having raised $63.5m in total through a $13.5m Series A led by Accel and a $50m Series B in March 2021 led by Insight Partners, with nothing published since. Named customers include Intuit, Autodesk, LendingClub and Starbucks.

Founded 2016 · Newark, California · trust3.ai