What is a data privacy vault?
A data privacy vault is an isolated, secured store for sensitive data such as PII, PHI and card numbers. Applications keep tokens instead of real values, and the vault holds the mapping between them. Only approved users and services can turn a token back into the real value, and every request is logged.
For example, an online pharmacy stores names, addresses and prescriptions. With a vault, the order system, analytics warehouse and support tools all hold tokens. When a pharmacist opens an order, the app asks the vault for the real name and prescription, and the vault returns them because her role allows it. A marketing analyst running a report on the same orders sees tokens.
How a data privacy vault works
- Detect sensitive values as data arrives, in structured fields and in free text.
- Tokenize each value and send the token to downstream systems.
- Store the real values and the token mapping in the isolated vault, encrypted.
- Detokenize by policy through an API, returning real values only to approved users and services.
- Log every request so teams can show who accessed what.
Stored in the CRM: <PERSON>tok_19bd</PERSON>, <EMAIL>tok_8f2a</EMAIL> Shown to pharmacist: Maria Lopez, maria.lopez@example.com
Data privacy vault vs encryption
| Data privacy vault | Encryption alone | |
|---|---|---|
| Where real data lives | In one isolated store | In every system, as ciphertext |
| Who can read it | Approved callers, checked against policy on each request | Anyone who holds the key |
| What downstream systems hold | Tokens | Encrypted values or, once decrypted, real values |
| Format | Tokens can keep the original format | Changes the format unless format-preserving encryption is used |
Why a vault matters for AI systems
AI systems make many copies of data. Prompts, model logs, vector stores and AI agent memory can all end up holding customer details. If those copies contain tokens instead of real values, a leak from any of them exposes tokens. Because the same value always maps to the same token, models can still follow who did what in a conversation or document.
Tokenization is pseudonymization, not anonymization. Tokenized data can still be personal data under GDPR and PHI under HIPAA, so the vault itself needs strong access control and audit.
A vault also makes privacy requests simpler. When a customer asks to be deleted, the team removes the real values from one place, and the tokens left in other systems no longer point to anyone.
How Protecto helps
Protecto Vault tokenizes sensitive data before it reaches models, AI agents and data stores, and reveals real values only to the roles and tools your policy allows. It is a headless API, so you can build it into your own auth, identity and access systems. Customers set how long Protecto retains sensitive data and when it is deleted. Protecto is SOC 2 Type II and ISO 27001 certified. See Protecto Vault.
Frequently asked questions
What is the difference between a data vault and a data privacy vault?
Data Vault is a data warehouse modeling method for storing history. A data privacy vault is a security design that isolates sensitive values and replaces them with tokens everywhere else. Read more in data privacy vault.
Is tokenized data anonymous?
No. Tokenization is pseudonymization, because approved users can map tokens back to real values. Laws such as GDPR still treat tokenized data as personal data.
Where does a data privacy vault run?
Some vaults run as a hosted service and others run in the customer’s own cloud or data center. Data residency rules often decide which option a team picks.
Related terms
Tokenization · Detokenization · Controlled unmasking · Pseudonymization · Data encryption · Format-preserving encryption