Data Anonymizer — CSV PII Removal

Paste a CSV into the Data Anonymizer, and it reads the header row to auto-detect columns that look like PII. Name, email, phone, SSN, date of birth and similar fields are toggled on by default; columns that look like IDs or categories are left alone. Click any toggle to change that judgment for a column, pick a strategy, and the anonymised CSV updates in real time.

Three strategies cover the most common compliance and testing workflows. Redact replaces every cell value with a placeholder tag, so a name becomes [NAME] and an email becomes [EMAIL]. Hash applies an 8-character hex digest (djb2) so the field is unreadable but the same input always maps to the same token, which preserves relational joins. Fake substitutes a plausible synthetic value drawn from pools of invented names, fiction-block phone numbers and unreachable email addresses, so the output still looks like a real dataset to a user interface under test.

Step by Step: Anonymising the Sample Customer Table

1. Click Load Sample. The tool loads a 5-row CSV with columns id, name, email, phone, age and city.
2. The column panel opens automatically. name, email and phone are toggled on (detected as PII);id, age and city are off.
3. With Redact selected, the anonymised CSV replaces Alice Thornton with [NAME], alice.thornton@mailexample.test with [EMAIL] and 555-0142 with [PHONE] across all five rows.
4. Switch to Fake for the name column only: each row receives a synthetic name from the fiction pool while email and phone stay redacted. Column-level strategy overrides the default.
5. Press Download to save anonymized-data.csv.

Edge Cases: What Gets Detected and Where It Stops

• PII detection is heuristic: column names containing name, email, phone, ssn, dob, address, zip, passport or credit (case-insensitive, separators stripped) are auto-toggled. Any column not matching that list starts as off.
• Fake values are not unique. The pool cycles modulo 10, so a table with more than 10 rows will repeat fake names and emails. For uniqueness across large datasets, use Hash instead.
• The Hash strategy applies djb2, an 8-hex-char non-cryptographic digest. It is deterministic (same input, same token) and useful for referential integrity, but not suitable for security-grade pseudonymisation. A dataset that requires GDPR pseudonymisation for legal purposes needs a keyed cryptographic function.
• Cell content is anonymised; column names are preserved. If your column is literally named name and that itself is sensitive, rename it before uploading.
• Input is capped only by browser memory. A 10,000-row CSV processes in under a second on a modern machine, because the entire computation is a string walk with no server round-trip.