Datasets:

aitrainingdatabroker
/

data-compliance-frameworks

license: cc-by-4.0
language:
  - en
pretty_name: Data Compliance Frameworks
size_categories:
  - n<1K
tags:
  - compliance
  - privacy
  - hipaa
  - gdpr
  - ccpa
  - de-identification
  - data-licensing
  - ai-training-data
  - regulations
configs:
  - config_name: frameworks
    data_files: data/frameworks.jsonl
  - config_name: elements
    data_files: data/elements.jsonl
  - config_name: crosswalk
    data_files: data/crosswalk.jsonl
  - config_name: obligations
    data_files: data/obligations.jsonl

Data Compliance Frameworks

A working aid for removing personal information and asking the right questions before licensing company data to an AI lab. It is organised around the rules that come up most often: HIPAA, 42 CFR Part 2, CCPA and CPRA, other US state privacy laws, GDPR, UK GDPR, PIPEDA and Quebec Law 25, GLBA, FCRA, COPPA, PCI DSS, CAN-SPAM, the Fair Housing Act, the Driver's Privacy Protection Act, ITAR and EAR, CUI and DFARS, GINA, 21 CFR Part 11, attorney-client privilege, contractual confidentiality and trade secret protection.

It does not make anyone compliant. It summarises what each rule is about so you can find the data it touches, clean it, and brief your counsel. Whether a rule applies depends on your data and where the people in it live.

Four tables

frameworks (21 rows): one row per rule or standard.

Field	Meaning
`framework`, `type`, `jurisdiction`, `oversight`	what it is, where it applies, who oversees it
`covers`	who it applies to
`protects`	what information it protects
`key_duties`	list of the main duties in plain words
`cleaning_implications`	what it means for removing data before licensing
`keywords`	terms that signal this rule in your records
`questions_for_counsel`	questions worth asking
`related_industries`	industries where it often comes up

elements (45 rows): the data elements three laws treat as personal: HIPAA's 18 Safe Harbor identifiers and its no-actual-knowledge condition, CCPA and CPRA's categories of personal information and sensitive personal information, and GDPR's personal data identifiers and special categories. Fields: law, source, category, element, note, placeholder_when_cleaning.

crosswalk (17 rows): one row per kind of data (name, email, phone, address, precise dates, government IDs, health, genetic, biometric, IP, device IDs, location, photos, voice, employment, race and religion) showing how HIPAA, CCPA and CPRA, and GDPR each treat it, the placeholder to use when cleaning, and keywords to scan for.

obligations (26 rows): methods, principles and rights for HIPAA (Safe Harbor, Expert Determination), CCPA and CPRA (deidentified information and consumer rights) and GDPR and UK GDPR (principles, pseudonymisation versus anonymisation, data subject rights).

Load it

from datasets import load_dataset
cw = load_dataset("aitrainingdatabroker/data-compliance-frameworks", "crosswalk", split="train")
for r in cw:
    print(r["element"], "|", r["hipaa"], "|", r["ccpa_cpra"], "|", r["gdpr"])

How to use it well

Start with crosswalk to see where the three big laws agree and differ for each kind of data.
Scan your records with the keywords, then read cleaning_implications for each rule that shows up.
Take questions_for_counsel to your counsel, and where a rule requires a qualified expert to sign off (for example HIPAA's Expert Determination), use one.
Clean the data, review a sample by hand, and approve it before it leaves your control.

Limits

One broker's summary, written in October 2026. Laws change and this is not a substitute for their current text.
It simplifies. Each rule has exceptions, thresholds and definitions that decide whether it applies.
Keywords are English strings and will miss other languages and misspellings.
PCI DSS is an industry standard and contractual confidentiality is a contract matter, not law. They are included because they limit what you can license.
Not legal advice and not a certification. Using this data does not make any dataset compliant or de-identified.

build.py rebuilds all four files from the source lists. A companion set, Data Compliance by Industry, maps the same rules to 21 industries and 170 industry keywords.

License and credit

CC BY 4.0. Please credit AI Training Data Broker and link to https://aitrainingdatabroker.com/de-identification/ when you share or build on it.

About

AI Training Data Broker is a broker and technical advisor for companies that sell or license their data to AI labs. It guides the technical side, including export, format, de-identification and compliance, and publishes free checklists on removing personal information and keeping data within the privacy rules that apply.