Google Cloud Sensitive Data Protection (formerly Cloud DLP)

• Sensitive Data Protection (formerly Cloud DLP) is a fully managed service designed to help discover, classify, and protect the most sensitive data inside and outside Google Cloud.
• Sensitive Data Protection helps inspect the data to gain valuable insights and make informed decisions to secure your data.
• It effectively reduces data risk with de-identification methods like masking, tokenization, and encryption.
• It seamlessly inspects and transforms structured and unstructured data.
• Sensitive Data Protection is deeply integrated into Security Command Center Enterprise risk engine for continuous data monitoring, high-value asset identification, vulnerability analysis, and attack scenario simulation.
• Supports data in Google Cloud (BigQuery, Cloud Storage, Cloud SQL, Datastore), multicloud (Amazon S3, Azure Blob Storage), and external sources via hybrid jobs.

Sensitive Data Protection Services

• Sensitive Data Protection is a family of services comprising three core capabilities:
  • Discovery – Automatically discovers, classifies, and profiles data across organizations, folders, or projects.
  • Inspection – Detects and classifies sensitive data in content, storage, and images.
  • De-identification – Transforms sensitive data to reduce risk while retaining utility.

Sensitive Data Protection Discovery Service

• The discovery service continuously and automatically discovers, classifies, and profiles data across an organization.
• Discovery helps understand the location and nature of data, including shadow data that might not be subject to proper governance.
• Discovery generates data profiles at project, table, and column levels providing:
  • Data classifications (infoTypes detected)
  • Sensitivity levels and data risk levels
  • Data size and shape
  • Data security posture
  • Estimated null proportion and uniqueness (at column level)
• Supported Data Sources:
  • BigQuery (GA since April 2022) – including BigLake tables
  • Cloud SQL (GA since March 2024)
  • Cloud Storage (GA since June 2024)
  • Amazon S3 (Sept 2024) – requires Security Command Center Enterprise with AWS connector
  • Azure Blob Storage (April 2025) – requires Security Command Center Enterprise with Azure connector
  • Vertex AI datasets (GA Feb 2025) – profiles training and tuning data for sensitivity
  • Vertex AI tuning jobs (Feb 2026)
• Discovery can automatically attach tags to resources based on calculated data sensitivity, enabling IAM conditions to grant or deny access based on data sensitivity.
• Discovery can detect unencrypted secrets (passwords, authentication tokens) in Cloud Functions and Cloud Run environment variables, reporting findings to Security Command Center.
• Discovery findings can be published to Security Command Center, Dataplex Universal Catalog (Knowledge Catalog), and BigQuery.
• Supports subscription pricing mode for predictable costs regardless of data growth.
• Scan configurations are manageable via Terraform (since June 2024).

Data Security Posture Management (DSPM)

• DSPM provides a data-centric view of Google Cloud security, integrated with Security Command Center.
• DSPM lets you continuously identify and reduce data risk by understanding what sensitive data you have, where it is stored, and if its use aligns with security and compliance requirements.
• Starts with a data map providing a birds-eye view of data across your environment, its sensitivity level, and default security posture.
• Sensitive Data Protection discovery feeds sensitivity signals into DSPM for default data risk assessment.

Sensitive Data Protection Actions

• An action occurs after a Sensitive Data Protection job completes successfully or, in the case of emails, on error.
• Supported action types:
  • Save findings to BigQuery (inspection and risk jobs)
  • Save findings to Cloud Storage (inspection jobs – added Aug 2025)
  • Publish to Pub/Sub (inspection and risk jobs)
  • Publish to Security Command Center (risk jobs and discovery)
  • Add Dataplex Catalog aspects based on insights from data profiles (replaced Data Catalog integration)
  • Publish to Google Cloud’s operations suite (risk jobs)
  • Notify by email (inspection and risk jobs)
  • Send inspection results to Dataplex Universal Catalog (Knowledge Catalog) as aspects (Sept 2025)

Sensitive Data Protection Key Concepts

• Classification is the process to inspect the data and know what data we have, how sensitive it is, and the likelihood.
• De-identification is the process of removing identifying information from data.
• De-identification techniques supported:
  • Redaction: Deletes all or part of a detected sensitive value.
  • Replacement: Replaces a detected sensitive value with a specified surrogate value.
  • Masking: Replaces a number of characters of a sensitive value with a specified surrogate character, such as a hash (#) or asterisk (*).
  • Pseudonymization/Tokenization: Replaces sensitive data values with cryptographically generated tokens. Supports:
    • CryptoReplaceFfxFpeConfig – Format-preserving encryption
    • CryptoDeterministicConfig – Deterministic encryption for consistent tokenization
    • CryptoHashConfig – One-way cryptographic hashing
  • Built-in Tokenization (Jan 2025): Serverless API endpoints for on-the-fly tokenization without managing third-party deployments, hardware, or VMs. Fully regionalized for compliance.
  • Generalization: Abstracts a distinguishing value into a more general, less distinguishing value to preserve data utility while reducing identifiability.
  • Bucketing: “Generalizes” a sensitive value by replacing it with a range of values. (For example, replacing a specific age with an age range.)
  • Date shifting: Shifts sensitive date values by a random amount of time.
  • Time extraction: Extracts or preserves specified portions of date and time values.
  • Dictionary replacement: Replaces each detected sensitive value with a random value from a provided word list.

Sensitive Data Protection InfoTypes

• Sensitive Data Protection uses information types — or infoTypes — to define what it scans for.
• An infoType is a type of sensitive data, such as a name, email address, telephone number, identification number, credit card number, and so on.
• An infoType detector is the corresponding detection mechanism that matches an infoType’s matching criteria.
• InfoType detectors are used in scan configurations to determine what to inspect for and how to transform findings.
• Supported infoType detector types:
  • Built-in infoType detectors — specified by name, include country-specific and globally applicable data types. Categories include:
    • PII (names, addresses, phone numbers, dates of birth)
    • Financial (credit cards, bank accounts, tax IDs)
    • Credentials & secrets (API keys, passwords, auth tokens, encryption keys)
    • Medical data (medical records, health info)
    • Government IDs (passports, driver’s licenses, national IDs)
    • Document type classifiers (finance, legal, medical, R&D, source code)
    • Image/Object infoTypes (faces, signatures, passports, photo IDs, barcodes, license plates, persons, whiteboards)
    • Image safety classifiers (violence, sexually explicit/suggestive content)
  • General infoType detectors — broader categories (e.g., CREDIT_CARD_DATA, MEDICAL_DATA, GOVERNMENT_ID) that cover multiple specific infoTypes.
  • Custom infoType detectors, defined by you:
    • Small custom dictionary detectors — simple word lists, ideal for several tens of thousands of words or phrases.
    • Large custom dictionary detectors — generated from large lists stored in Cloud Storage or BigQuery, ideal for up to tens of millions of words.
    • Regular expressions (regex) detectors — detect matches based on regex patterns.
    • Custom metadata label detectors (March 2026) — detect specific client-provided metadata labels.
• Inspection rules to fine-tune scan results:
  • Exclusion rules — decrease findings (including ExcludeByHotword for column-name-based exclusion and image-based exclusion rules).
  • Hotword rules — increase quantity or change likelihood of findings.
  • Adjustment rules (GA Feb 2026) — customize likelihood based on context, supporting both text and image operations.
• InfoType versioning — control which detection model to use via InfoType.version: latest, stable, or legacy.
• Likelihood levels used for confidence scoring:
  • LIKELIHOOD_UNSPECIFIED: Default value; same as POSSIBLE.
  • VERY_UNLIKELY
  • UNLIKELY
  • POSSIBLE
  • LIKELY
  • VERY_LIKELY

Classification and De-identification

• Sensitive Data Protection can classify, redact, and de-identify sensitive data contained in text-based content, images, and content stored in Google Cloud storage repositories.
• Text Classification and De-identification
  • Text Classification returns classification findings.
  • Text Redaction produces output with sensitive data removed using a placeholder.
  • Conversational content inspection (June 2026) — supports inspecting and de-identifying conversational content via the Conversation ContentItem type.
  • Batched content inspection (June 2026) — supports inspecting and de-identifying batched content via BatchContentItem.
• Image Classification and Redaction
  • Uses Optical Character Recognition (OCR) technology to recognize text prior to classification.
  • Inspection — inspects base64-encoded images for specified infoTypes, returns detected InfoTypes with pixel coordinates and bounding boxes.
  • Redaction — masks sensitive data findings with opaque rectangles (configurable color).
  • Object detection and redaction (July 2025) — detects and redacts sensitive objects (barcodes, license plates, persons, whiteboards) in images.
• Storage Classification
  • Scans data stored in Cloud Storage, Datastore, BigQuery, and Cloud SQL.
  • Supports scanning of binary, text, image, Microsoft Word, PDF, Apache Avro files, and archive files.
  • Detects sensitive data in headers and footers of rich document types.
  • Unrecognized file types are scanned as binary files.
• Date Handling — if dates are considered PII:
  • Using generalization or bucketing (can reduce date utility)
  • Using date shifting which randomly shifts dates but preserves sequence and duration.

Re-identification Risk Analysis

• Risk analysis is the process of analyzing sensitive data to find properties that might increase the risk of subjects being identified.
• Risk analysis methods can be used before de-identification to help determine an effective strategy, or after de-identification to monitor for changes or outliers.
• Supported risk metrics:
  • k-anonymity — measures whether individuals can be distinguished from at least k-1 other individuals.
  • l-diversity — extends k-anonymity by ensuring diversity of sensitive values within each equivalence class.
  • k-map estimation — estimates re-identification risk based on quasi-identifiers.
  • δ-presence (delta-presence) — risk metric for when membership in the dataset is itself sensitive.

Sensitive Data Protection Templates

• Templates help decouple configuration information from the implementation of requests.
• Templates provide a robust way to manage large-scale rollouts.
• Two types of templates:
  • Inspection Templates: Templates for saving configuration for inspection scan jobs, including what predefined or custom detectors to use.
  • De-identification Templates: Templates for saving configuration for de-identification jobs, including infoType and structured dataset transformations.
• Discovery scans can be configured to reprofile data when the inspection template changes (since Feb 2024).

AI and Generative AI Protection

• Sensitive Data Protection extends to AI workloads to help organizations safely unlock data value at every stage of the AI journey.
• Vertex AI Discovery (GA Feb 2025) — discovers and profiles sensitivity of training and tuning data in Vertex AI datasets.
• Context-aware inspection — understands data context even within images and rich documents for effective AI agent data access control.
• Fine-grained data minimization for:
  • AI model training and fine-tuning data preparation
  • Runtime protection for chat, data collection, and generative AI prompts/responses
  • Ensuring adherence to regulations and internal policies
• Integrated with AI Protection framework for security posture management of AI workloads.

Integration with Security Command Center

• Sensitive Data Protection is deeply integrated into Security Command Center Enterprise risk engine.
• Publishes the following finding types:
  • Data sensitivity and Data risk — calculated sensitivity and risk levels of profiled tables.
  • PUBLIC_SENSITIVE_DATA — publicly accessible resources with sensitive data.
  • SECRETS_IN_STORAGE — secrets detected in environment variables.
  • SENSITIVE_DATA_CMEK_DISABLED — sensitive data without customer-managed encryption.
• Security Command Center can automatically prioritize resources based on data sensitivity.
• Discovery is included in Security Command Center Enterprise tier (since Nov 2024).

Regional Endpoints and Data Residency

• Regional endpoints (GA Aug 2024) keep data at rest, in use, and in transit within a specified region.
• Available in 35+ regions globally including multi-regions (asia, europe, us).
• Content methods process data synchronously in memory — data is not stored on Google Cloud.
• Storage methods process data in the same region where it resides.

Compliance and Security

• Certifications: ISO/IEC 27001, ISO/IEC 27017, ISO/IEC 27018, PCI DSS, HIPAA BAA, MTCS SS 584.
• Supports GDPR compliance with built-in and custom infoType detectors applicable to EU data.
• Data encryption in transit and at rest follows Google Cloud encryption standards.

Reference

• Sensitive Data Protection Documentation
• Sensitive Data Protection Overview
• Discovery and Inspection
• Data Profiles
• InfoType Detector Reference
• Release Notes