Privacy Engineering

What is privacy engineering?

Privacy engineering is the practice of building privacy protections into systems, products and data flows as technical requirements, instead of relying on policy documents alone. Privacy engineers turn legal and policy rules into design choices: what data a feature collects, how it is masked, who can see it and when it is deleted.

For example, a product team adds an AI summary to recorded support calls. The privacy engineer maps the flow: the call transcript goes to a model, the summary is saved in the CRM and every request lands in application logs. She sets four rules. Names and card numbers are tokenized before the transcript reaches the model. Summaries stay tokenized in the CRM. Only the support rep assigned to the case can unmask them. Raw transcripts are deleted after 30 days.

Core practices

  • Data mapping. Trace where personal data enters, moves, is stored and leaves a system.
  • Privacy threat modeling. Look for ways data could be linked, identified or disclosed. LINDDUN is a widely used method.
  • Data minimization. Collect and keep only the fields a feature needs.
  • De-identification. Mask, tokenize or aggregate data so systems work without raw identities.
  • Access control. Grant the least access each person, service or AI agent needs, and reveal real values by role.
  • Retention and deletion. Automate deletion schedules and data subject requests across every copy.

Frameworks privacy engineers use

NIST describes three privacy engineering objectives. Predictability means people and operators can rely on how a system handles data. Manageability means data can be changed, deleted and controlled at a granular level. Disassociability means data can be processed without linking it to people more than needed. Teams also use the NIST Privacy Framework, ISO/IEC 27550 on privacy engineering for system life cycles, and the privacy by design principles.

Privacy engineering vs privacy compliance

Privacy compliance Privacy engineering
Main question What does the law and policy require? How does the system enforce it?
Typical output Policies, notices, assessments, records of processing Data flows, masking, access rules, deletion jobs, tests
Owner Legal and privacy teams Engineering, platform and security teams

The two depend on each other. A policy no system enforces is a gap, and a control with no policy behind it is guesswork.

Why privacy engineering matters for AI systems

AI adds data flows that older privacy designs never planned for: prompts, retrieved context in RAG, embeddings in vector stores, AI agent tool calls, Model Context Protocol (MCP) servers and long-lived logs. Each new AI feature tends to rebuild masking and access logic from scratch, and the copies drift apart. Privacy engineering for AI means one consistent set of controls across all of those paths.

How Protecto helps

Protecto gives privacy and platform engineers one place to set and enforce data policy for AI, so each team does not rebuild it. Protecto detects sensitive data, tokenizes it consistently, reveals real values by role and context, and reports on how policies are enforced. It runs in data pipelines through its API, a Snowflake UDF, Databricks, Kafka or Spark. See secure AI data pipelines.

Frequently asked questions

What does a privacy engineer do?

A privacy engineer turns privacy laws and policies into technical controls. That includes mapping data flows, designing masking and access rules, automating deletion and testing that systems handle personal data as intended.

What is the difference between privacy engineering and privacy by design?

Privacy by design is a set of principles that says privacy should be built in from the start. Privacy engineering is the hands-on work that puts those principles into code, architecture and operations.

Why do AI systems need privacy engineering?

AI moves personal data through prompts, retrieval, vector stores, AI agent tool calls and logs. Without designed controls, sensitive values spread to places no policy covers.

Related terms

Privacy by design · NIST Privacy Framework · Data minimization · Data mapping · Privacy impact assessment

Book a demo

See Protecto on your own data

Book a demo to see how Protecto finds and masks sensitive data before it reaches any AI model.

New: Protecto now runs inside LiteLLM. Control what sensitive data your models and tools see.
Read the launch blog