# Trust3 AI Data Governance

> Data and AI access governance platform from the creators of Apache Ranger: sensitive data discovery, plus access, masking and row-filter policies pushed down into Snowflake, Databricks, BigQuery and 50+ sources, and an inventory of AI agents with trust scores and EU AI Act checks.

Source: https://toolsfordata.com/tools/trust3-ai/ · Updated 2026-11-19

[Website](https://trust3.ai/) · [Docs](https://docs.privacera.com/) · [Pricing](https://aws.amazon.com/marketplace/pp/prodview-p3z3wy2ev3tgq)

Best for: security and data platform teams in regulated industries that need one Ranger-based policy layer across multi-cloud and on-premises data, and want to govern which data their AI agents can reach.

## Overview

- Core Strength: Apache Ranger policies pushed down into 50+ sources, plus governance of AI agents
- Lineage Depth: Partial — automated, table
- Catalog Curation: Partial — custom fields, propagation
- Stewardship Workflows: Partial — review/approval
- Policy & Classification: Full — sensitive-data discovery, policy, enforcement, masking
- Compliance Depth: Partial — audit trail, reporting, regulatory mapping, subject rights
- Marketplace & Data Products: Partial — data products, data mesh
- Built-in Data Quality: Not supported
- AI & Agent Interface: Partial — MCP, chat
- AI Governance: Full — AI inventory, AI lineage, AI regulation, model health…
- Developer Surface: Partial — API, Python SDK
- Deployment Model: Partial — SaaS, self-hosted, hybrid
- Security & Access: Full — SSO, SCIM, RBAC, audit log…
- Notable Caveat: Not a catalog: no glossary, certification or data quality, from $100,000 a year

## What it's used for

Trust3 AI, Privacera's name since March 2026, is used to set one set of Apache Ranger-based access, masking and row-filter policies and push them into Snowflake, Databricks, BigQuery, S3 and 50+ other sources. A newer module governs the AI agents teams build.

## Classification

Trust3 AI discovers sensitive data with regex, dictionaries and models, with a tag review queue. Ranger policies with row filters, masking, deny rules and validity periods are enforced by plugins or translated into native controls.

## AI governance

Trust3 AI finds AI agents on Databricks, Copilot Studio and Azure Foundry, gives each a Trust Score from 1 to 10, checks EU AI Act policy packs and runs a legal and compliance approval chain.

## Catalog

A light catalog and data products view shows resources with their tags and who can access them. There's no glossary, certification or rich lineage.

## Pricing

Trust3 AI doesn't publish prices. On AWS Marketplace, the Data Access Governance Starter pack lists at $100,000 for 12 months, plus $200 per extra unit.

## Deployment

Trust3 AI runs fully hosted, with a hosted control plane and connectors in your own Kubernetes cluster (the recommended option), or fully self-managed on AWS EKS.

## Security

Trust3 AI supports SAML SSO, SCIM user sync, portal roles and audit logs, with SOC 2 Type II.

## Integrations

Trust3 AI enforces access policies on Snowflake, Databricks, BigQuery and Redshift and on data lakes in S3 and Azure, and syncs tags with Collibra and Microsoft Purview.

## Review

Strengths:
- Pushes policies into Snowflake, Unity Catalog, BigQuery and Redshift as native controls, or enforces them with Ranger plugins on Spark, EMR and Hive.
- Built by the creators of Apache Ranger, so Ranger estates can move over without rewriting policies.
- Governs data access and AI agents in one product.

Caveats:
- Not a catalog: no glossary, certification, collaboration or data quality, and lineage covers only GCS and BigQuery.
- The lowest listed price is $100,000 a year on AWS Marketplace, with no free tier.
- Two doc sites and portals: one for data access, one for AI governance.

Recommendation: Choose Trust3 AI if you run Apache Ranger or need one policy layer across cloud and on-premises data. For access requests and data products, consider Immuta.

## Facts

- Privacera renamed itself Trust3 AI on March 9, 2026.
- Trust3 AI launched MCP Security, a runtime control layer for MCP traffic, in May 2026.
- PrivaceraCloud has been on AWS Marketplace since June 2021.

## Pricing

- Model: custom
- Entry team price: Custom
- Entry enterprise price: $100,000/year
- Source: https://aws.amazon.com/marketplace/pp/prodview-p3z3wy2ev3tgq (checked 2026-11-12)

## Features

### Data Catalog

- Automated Metadata Harvesting: Yes — Connectors pull the resource hierarchy (databases, schemas, tables, columns, volumes, buckets) from each connected platform into the catalog without manual entry. ([source](https://docs.privacera.com/user-guide/data-governance/data-catalog.html))
- Search & Discovery: No — Catalog search is a name filter on the navigation tree within one data source; there is no faceted search by owner, tag or certification. ([source](https://docs.privacera.com/user-guide/data-governance/data-catalog.html))
- AI-Powered Search: No — The GIA assistant answers plain-language questions about the AI agent inventory, not about data assets in the catalog; the docs show no natural-language catalog search. ([source](https://docs.trust3.ai/platform-concepts/gia.html))
- Usage & Popularity Statistics: No — Access audits log every query, but the catalog shows no per-asset query counts, users or popularity ranking. ([source](https://docs.privacera.com/user-guide/data-governance/data-catalog.html))
- Custom Fields & Templates: Yes — Tag definitions carry custom attribute schemas with data types, required fields, validation rules and helper text, attached to chosen resource types. ([source](https://docs.privacera.com/user-guide/data-governance/tags-and-metadata.html))
- Asset Certification: No — The docs describe no Verified, Endorsed or Deprecated status for tables or other assets. ([source](https://docs.privacera.com/user-guide/data-governance/data-catalog.html))
- Business Glossary: No — There is no business glossary module; governance metadata is tags and tag attributes. ([source](https://docs.privacera.com/user-guide/data-governance/index.html))
- Metadata Propagation: Yes — Tags are inherited from parent resources in the catalog, and Discovery can propagate classification tags from scanned part files up to their parent folders. ([source](https://docs.privacera.com/get-started/discovery/config_part_file_tag_propagation.html))

### Data Lineage

- Table-Level Lineage: Yes — Discovery builds a resource lineage graph of source-to-destination events for GCS and BigQuery; every other Discovery connector lists lineage as not supported. ([source](https://docs.privacera.com/user-guide/discovery/advanced/solr_collection_retention.html))
- Column-Level Lineage: No — The docs describe no column-level lineage; Discovery lineage tracks resource movement only. ([source](https://docs.privacera.com/connectors/gcp-bigquery/discovery/index.html))
- End-to-End Lineage: No — Lineage does not reach BI dashboards or ingestion pipelines; Discovery lineage covers GCS and BigQuery resource movement only. ([source](https://docs.privacera.com/connectors/gcp-bigquery/discovery/index.html))
- Impact Analysis: No — The docs describe no downstream impact analysis for schema or table changes. ([source](https://docs.privacera.com/connectors/gcp-bigquery/discovery/index.html))
- Automated Lineage Extraction: Yes — Discovery builds lineage from GCS and BigQuery audit-log events read through a Google logging sink and Pub/Sub; this is the only automated lineage source. ([source](https://docs.privacera.com/get-started/discovery/prerequisites/prerequisites-google-cloud.html))
- OpenLineage Support: No — Neither the Privacera nor the Trust3 docs mention OpenLineage. ([source](https://docs.privacera.com/connectors/gcp-bigquery/discovery/index.html))

### Data Stewardship

- Critical Data Elements: No — The docs describe no way to register or monitor critical data elements. ([source](https://docs.privacera.com/user-guide/data-governance/index.html))
- Review & Approval Workflows: Yes — New AI agents go through a pre-development approval chain with Legal and Compliance reviews and a permanent record of who approved what; Discovery also has a review queue to accept or reject auto-applied tags. ([source](https://docs.trust3.ai/how-to-guides/pre-development-workflow.html))
- Governance Campaigns: No — Bulk Policy Updates edits many Data Zone policies at once, but there are no campaigns that assign owners or certify groups of assets. ([source](https://docs.privacera.com/user-guide/discovery/policies/compliance/bulk_policy_updates.html))
- Governance Scorecards: No — There are no questionnaire-based maturity scorecards; the Trust Score rates individual AI agents, not governance maturity of domains. ([source](https://docs.trust3.ai/platform-concepts/trust-score.html))
- Metadata Automation: Yes — Discovery real-time and scheduled scans tag newly found sensitive data automatically, and tag-based policies apply to those resources as soon as the tag lands. ([source](https://docs.privacera.com/user-guide/discovery/scanning/ways/realtime-scanning.html))

### Data Classification

- Sensitive Data Discovery: Yes — Discovery classifies data automatically with regex patterns, dictionaries and heuristic models (credit card with Luhn check, SSN, DOB, national IDs and more), then lets users review the results. ([source](https://docs.privacera.com/user-guide/discovery/scanning/classification/models.html))
- Data Policy: Yes — Apache Ranger-based resource, tag and attribute-based policies control access, with row-level filters, column masking, deny and exclude rules, security zones and time-bound validity. ([source](https://docs.privacera.com/user-guide/access-management/access_policy_definition/index.html))
- Native Policy Enforcement: Yes — PolicySync connectors translate central policies into native permissions, masking and row filters in Snowflake, Databricks Unity Catalog, BigQuery, Redshift and other platforms, which then enforce them. ([source](https://docs.privacera.com/resources/design/access-management/integrations/privacera_policysync.html))
- Data Masking: Yes — Policies apply dynamic column masking (hash, nullify, show last 4 and others), and the encryption module adds format-preserving encryption and static masking schemes. ([source](https://docs.privacera.com/user-guide/access-management/access_policy_definition/column_masking.html))

### AI Features

- AI Chat Assistant: Yes — GIA answers plain-language governance questions over the live agent inventory, explains trust score drops and fills agent registration forms. ([source](https://docs.trust3.ai/platform-concepts/gia.html))
- AI Documentation Generation: No — The docs describe no LLM-written descriptions for tables or columns; GIA drafts agent registration forms only. ([source](https://docs.trust3.ai/platform-concepts/gia.html))
- Agentic Metadata Enrichment: Unknown — The March 2026 launch talks about Trust Agents for discovery and policy creation, but the docs show no agent that enriches catalog metadata. ([source](https://privacera.com/newsroom/press-releases/meet-trust3-ai/))

### Data Compliance

- Audit Trail: Yes — Access audits log every allowed or denied data access across connected platforms, and admin audits log every policy, role and tag change. ([source](https://docs.privacera.com/user-guide/access-management/audits/access_audits.html))
- Compliance Reporting: Yes — Discovery has summary and drill-down reports on classified data, and the AI governance module produces timestamped evidence for audits of agents. ([source](https://docs.privacera.com/user-guide/discovery/reports/generating/summary.html))
- Regulatory Framework Mapping: Yes — Pre-built policy packs map agent requirements to EU AI Act articles, NERC CIP, FERC and internal AI governance, and GIA can turn a compliance document into policy rules. ([source](https://docs.trust3.ai/platform-concepts/policies.html))
- Data Activity Monitoring: Yes — Access audits collect and normalize actual query events from each data platform, and Posture Manager flags unusual access patterns. ([source](https://docs.privacera.com/user-guide/access-management/audits/access_audits.html))
- ROPA: No — There is no GDPR Article 30 processing register. ([source](https://docs.privacera.com/user-guide/discovery/policies/compliance/workflow.html))
- Data Retention Policies: No — There are no time-based retention rules for source data; the Expunge policy deletes records that match a lookup file, not records past an age. ([source](https://docs.privacera.com/user-guide/discovery/policies/compliance/expunge_policy.html))
- Data Subject Rights: Yes — The Right to Privacy policy masks a data subject's records matched from a lookup file and the Expunge policy removes them, on S3, ADLS, Snowflake, Redshift and JDBC sources. ([source](https://docs.privacera.com/user-guide/discovery/policies/compliance/rtp_policy.html))

### Collaboration

- Alert Notifications: No — Slack, Teams and webhook alerts come from Grafana and cover connector and runtime health, not governance events like ownership or access changes. ([source](https://docs.privacera.com/user-guide/runtime-plane/alert_notification.html))
- Discussions & Annotations: No — The docs describe no comments or threaded discussions on assets. ([source](https://docs.privacera.com/user-guide/data-governance/data-catalog.html))
- Ticketing Integrations: No — The docs describe no Jira or ServiceNow ticket creation; AI policy violations open issues inside Trust3 only. ([source](https://docs.trust3.ai/platform-concepts/policies.html))

### Data Marketplace

- Self-Service Access Requests: Unknown — The stewardship page mentions an approval workflow and one-click provisioning, but the docs show no Request Access button for data consumers. ([source](https://privacera.com/products/governed-data-stewardship/))
- Data Products: Yes — Data Products group tables and columns from several platforms into a named, draft-or-published product that gets its own tag and Ranger role for policies; there are no SLAs or contracts. ([source](https://docs.privacera.com/user-guide/data-governance/data-products.html))
- Data Mesh Support: Yes — Security zones and delegated admins let domain teams own policies for their own resources under central rules, which Privacera sells as federated data stewardship. ([source](https://docs.privacera.com/user-guide/access-management/access_policy_definition/security_zones.html))
- Subscription Management: No — The docs describe no subscriptions to data products or change notifications for consumers. ([source](https://docs.privacera.com/user-guide/data-governance/data-products.html))

### Data Quality

- SQL Table Tests: No — There are no data quality rules or SQL assertions. ([source](https://docs.privacera.com/user-guide/data-governance/index.html))
- Data Health Scores: No — There are no quality scores for tables; the Trust Score rates AI agents' governance. ([source](https://docs.privacera.com/user-guide/data-governance/index.html))
- Anomaly Detection: No — Anomaly detection covers unusual access and agent behaviour, not data values or volumes. ([source](https://docs.privacera.com/user-guide/data-governance/index.html))
- Data Profiling: No — Discovery samples data to classify it but does not report column statistics like null rates or distributions. ([source](https://docs.privacera.com/user-guide/discovery/intro/features.html))
- Data Contracts: No — There are no producer-consumer data contracts. ([source](https://docs.privacera.com/user-guide/data-governance/index.html))

### AI Governance

- AI Asset Inventory: Yes — The inventory records each agent, tool, MCP server and model with platform, environment, owners, purpose and who can invoke it, from scheduled discovery, a form or the Python collector SDK. ([source](https://docs.trust3.ai/get-started/core-concepts.html))
- AI Lineage: Yes — Each agent's Lineage tab shows the datasets, tables, models, MCP servers, tools and knowledge sources it is built on; this is dependency lineage, not training-run lineage. ([source](https://docs.trust3.ai/get-started/core-concepts.html))
- AI Regulatory Compliance: Yes — EU AI Act policies check risk classification (Article 6), technical documentation (Article 11), human oversight (Article 14), training data provenance (Article 10) and EU database registration (Article 49). ([source](https://docs.trust3.ai/platform-concepts/policies.html))
- AI Model Health Score: Yes — Every agent gets a Trust Score from 1.0 to 10.0 with High, Medium, Low or Critical bands that drop with open violations, missing owners or unapproved models. ([source](https://docs.trust3.ai/platform-concepts/trust-score.html))
- AI Agent Access Governance: Yes — Runtime Data Access policies run in the collector on each agent request and can block or warn based on the classification of the data the agent is reaching and the destination type. ([source](https://docs.trust3.ai/platform-concepts/policies.html))

### Master Data Management

- Master Data Management: No — Trust3 AI has no master data management. ([source](https://docs.privacera.com/user-guide/data-governance/index.html))

### Workflows

- REST / GraphQL API: Yes — REST APIs cover Ranger policies, Discovery scans, tag and resource registration, encryption (PEG) and runtime plane management. ([source](https://docs.privacera.com/resources/api-reference/index.html))
- dbt Integration: No — Neither the Privacera nor the Trust3 docs mention dbt. ([source](https://docs.privacera.com/connectors/index.html))
- Python SDK: Yes — The pip package trust3-ai-collector-sdk publishes agents, tools, models and their relationships into the AI asset inventory; it covers AI assets, not the data catalog. ([source](https://docs.trust3.ai/ai-assets-collector/publish-inventory.html))
- Webhooks: No — Webhook contact points exist only for Grafana runtime health alerts, not for catalog or policy events. ([source](https://docs.privacera.com/user-guide/runtime-plane/alert_notification.html))

### Data Context Layer

- MCP Server: Yes — A hosted Trust3 MCP server gives Cursor, Claude Desktop and other clients read-only access to agent inventory, policies and trust scores with a scoped API key. ([source](https://docs.trust3.ai/how-to-guides/mcp-configuration.html))
- Knowledge Graph: Unknown — The AI inventory stores assets and typed relationships (CAN_INVOKE, USES_MODEL, HAS_ACCESS_TO) that GIA queries as an identity graph, but the docs do not describe a property graph of the whole data estate. ([source](https://docs.trust3.ai/ai-assets-collector/publish-inventory.html))
- Semantic Models: Unknown — Trust3 IQ was announced in 2025 with a universal semantic layer, but it is absent from the current site and docs. ([source](https://www.einpresswire.com/article/840907580/trust3-iq-from-trust3-ai-raises-the-bar-for-ai-accuracy-with-universal-semantic-layer-and-context-engine))
- Verified Queries: No — The GIA assistant answers governance questions about the agent inventory; there is no text-to-SQL and no store of approved queries. ([source](https://docs.trust3.ai/platform-concepts/gia.html))
- Semantic Layer Interop: No — The docs describe no import or export of metrics or semantic models; the Trust3 IQ semantic layer announced in 2025 is not in the current docs. ([source](https://docs.trust3.ai/get-started/core-concepts.html))
- Unstructured Data Catalog: No — Discovery scans .doc, .docx, .pdf and .txt files for sensitive data, but the Data Catalog lists only databases, schemas, tables and columns, and knowledge sources appear only as agent dependencies. ([source](https://docs.privacera.com/user-guide/discovery/sources/index.html))

### Security

- SSO: Yes — Portal login supports SAML SSO with Okta, Entra ID, OneLogin and Ping, plus OAuth and LDAP. ([source](https://docs.privacera.com/get-started/user-management/sso/index.html))
- RBAC: Yes — Portal roles (Account Admin, Policy Admin, Policy Auditor, User) control who manages users, policies, zones, tags and audits; the AI module has Admin, Compliance, Legal, Developer and Viewer roles. ([source](https://docs.privacera.com/user-guide/access-management/portal_roles_permissions.html))
- Audit Log: Yes — Admin audits record policy, role, group and tag changes, and login activity logs record each login with IP, method and result. ([source](https://docs.privacera.com/user-guide/access-management/audits/login_activity_logs.html))
- Compliance Certifications: Yes — The trust center states a SOC 2 Type II attestation; no ISO 27001 or FedRAMP is listed. ([source](https://privacera.com/trust-center/))
- SCIM Provisioning: Yes — UserSync accepts SCIM 2.0 push from Okta, Entra ID and OneLogin (the recommended method) or pulls from LDAP, AD, Okta and Entra. ([source](https://docs.privacera.com/user-guide/access-management/access_policy_definition/user_principals/usersync_guide.html))

### Deployment

- SaaS Deployment: Yes — In Trust3 AI Cloud the whole stack, connectors included, runs in Trust3's cloud and nothing is installed in the customer account. ([source](https://docs.privacera.com/user-guide/runtime-plane/deployment-options.html))
- Self-Hosted Deployment: Yes — Self-Managed runs the full stack, control plane included, in the customer's own AWS EKS account. ([source](https://docs.privacera.com/user-guide/runtime-plane/deployment-options.html))
- Hybrid Deployment: Yes — Partially Managed (D2P), the recommended option, keeps the control plane in Trust3's cloud while the runtime agent and connectors run in the customer's AWS, Azure or Google Cloud account. ([source](https://docs.privacera.com/user-guide/runtime-plane/deployment-options.html))
- Data Residency: Unknown — The hosted MCP endpoint is labelled NA cloud, but the docs do not list other regions to choose from for the SaaS control plane. ([source](https://docs.trust3.ai/how-to-guides/mcp-configuration.html))

### Open Source

- Open Source: No — The platform is proprietary; it builds on Apache Ranger, an Apache project, and the Apache-2.0 PAIG repo for GenAI guardrails has had no commits since August 2025. ([source](https://github.com/privacera/paig))

## Integrations

39 integrations, each sourced from the vendor docs.

- Authentication: AWS PrivateLink, Azure Key Vault, Google OAuth, LDAP, Microsoft Entra, Okta, Onelogin, OpenID, Ping, SAML
- Data Governance: AWS Glue Data Catalog, AWS Lake Formation, Collibra, Databricks Unity Catalog, Microsoft Purview
- Data Storage: Amazon S3, Azure Data Lake Storage, Google Cloud Storage
- Data Transformation: Amazon EMR, Apache Spark
- Data Warehouse: Amazon Redshift, Azure Synapse, Databricks SQL, Dremio, Google BigQuery, OpenText Analytics Database (Vertica), Oracle DB, Snowflake, Starburst
- Databases: Amazon DynamoDB, Google Cloud SQL, Microsoft SQL Server, MySQL, PostgreSQL
- Query Engine: Amazon Athena, Trino
- Streaming: Amazon SNS, Amazon SQS, Apache Flink

## Alternatives

- [Immuta](https://toolsfordata.com/tools/immuta/): The closest match: policy-as-code access control across Snowflake, Databricks, BigQuery and Redshift.
- [Microsoft Purview](https://toolsfordata.com/tools/microsoft-purview/): Consider Purview for data security and compliance on a Microsoft estate.
- [Databricks Unity Catalog](https://toolsfordata.com/tools/databricks-unity-catalog/): Consider Unity Catalog if most data is on Databricks: fine-grained access control is built in.
- [Snowflake Horizon](https://toolsfordata.com/tools/snowflake-horizon/): Consider Horizon if most data is on Snowflake: masking and policies are built in.
- [Apache Atlas](https://toolsfordata.com/tools/apache-atlas/): The free option from the same Apache family: tag-based policies and classification for Hadoop estates.
