Check your dataset before the model sees it. Nothing uploads.

Drop a CSV and find the sensitive fields — emails, IDs, dates of birth, health and financial columns — before you train a model, share the file, or publish the dataset. Entirely in your browser.

  • Per-column verdicts
  • Quasi-identifier warnings
  • Downloadable audit report
  • Free, no account, no uploads

Check a CSV dataset for sensitive fields

Drop a .csv file here, or click to choose one

Up to 5 MB. The file is read and scanned in your browser — it is never sent anywhere.

0 / 200,000 characters

Paste a CSV or drop a .csv file, then press Check dataset.

Do not paste real personal data into any website, this one included, without need — check a small extract, or a copy with the most sensitive values already removed.

Privacy you can verify, not just trust

Checking a dataset for personal data by uploading it to a website is a contradiction — the check itself would be the leak. This one can't: your CSV is parsed and scanned with plain JavaScript on your device, and the file never leaves the device — the only third-party requests are the site's own analytics, which count page views. Even the examples shown in the results are masked, and the downloaded report contains masked samples only. Don't take our word for it:

New cross-origin requests during use: 0
  1. Press F12 to open developer tools, then the Network tab.
  2. Press Load example and check the dataset above. Watch: apart from the site's analytics, nothing new is requested — no uploads, and no data ever leaves your device.
  3. Extra credit: disconnect your internet after this page loads — checking still works.

How to check a dataset for sensitive data before training

Models remember what they are trained on. A customer email, a date of birth or a salary figure that slips into a training set can surface later in a model's output, in a log, or in a dataset someone else downloads. The check takes a minute and belongs before any of that happens — before fine-tuning, before sharing a file with a colleague, and before publishing a dataset on a hub.

To use it:

  1. Drop a .csv file (up to 5 MB) or paste the CSV text with its header row. Comma, semicolon, tab or pipe separated — the delimiter is detected from the first line.
  2. Press Check dataset. The checker samples up to 1,000 rows evenly across the file, then matches every column two ways: by its name (email, date_of_birth, salary, diagnosis) and by the shape of its values (email addresses, phone numbers, card numbers checked with the Luhn algorithm, IBANs checked with mod-97, IP addresses, dates, postcodes).
  3. Work through the Sensitive columns first. Each finding shows what was detected, how much of the column matches, a masked example, and a recommended treatment — remove, hash, generalise or pseudonymise.
  4. Check the quasi-identifier warning. Postcode plus age plus gender can single out a person even after the direct identifiers are gone. Generalise or drop at least one of each pair.
  5. Download the report (Markdown or JSON) and keep it with the dataset as your audit trail.

What makes this checker different

The free tools that rank for this search are scrubbers, not checkers: PrivHaven scrubs CSVs in the browser but caps its free tier at 10,000 rows, keeps its reproducible report behind Pro, and does not detect names or addresses at all; the Gera Tools Data Anonymizer replaces values with fakes but produces no audit report and no per-column risk verdict; the Unfckr CSV PII Scrubber masks, hashes or drops columns but likewise covers only emails, phones, SSNs and cards, with no names, health or financial columns and no report. The developer options — Microsoft Presidio, privacy-scan-ml and releaseguard — are libraries and command-line tools that need installing and a Python environment, and PII Crawler is a $497 desktop app. This page is the free, no-account, no-upload checker built for the AI-training question specifically: per-column verdicts across 24 sensitive types, quasi-identifier warnings, masked samples, and a downloadable audit report you can keep with the dataset.

When is this the right check?

Whenever a dataset is about to leave your machine with people in it: before fine-tuning or training a model, before sharing a file with a colleague or a contractor, before publishing a dataset, and before pasting rows into an AI chat. It is also the fast first pass for GDPR and EU AI Act dataset governance — a record of what was checked and what was flagged, in the report you download.

Honest limitations

Pattern-based, not a guarantee

Detection matches column names and value shapes. A clean result means no sensitive pattern was found in the sampled rows — it does not prove the dataset is safe. Spot-check the data yourself before publishing or training on it.

Names come from headers only

Values alone cannot reliably identify a name, so names are detected from column names such as first_name or full_name. A column called "col_7" full of names will not be flagged — rename columns honestly before scanning.

No free-text entity recognition

A long notes column full of names, addresses and phone numbers in running text is not scanned as entities. If a column holds free text about people, treat it as sensitive and review it by eye.

Only a sample is scanned

Up to 1,000 rows, sampled evenly. A sensitive value that appears only in row 50,000 will not be seen. For a full check, scan the file in extracts.

It checks, it does not clean

This tool flags sensitive columns and tells you what to do with each one — it does not redact, mask or rewrite your file. Apply the recommendations in your own pipeline, then re-check the cleaned file.

Not legal advice

The verdicts and recommendations are a practical first pass, not a compliance opinion. GDPR, the EU AI Act and sector rules have their own requirements — check them, or ask a qualified adviser, before publishing personal data.

Frequently asked questions

Is a clean result a guarantee my dataset is safe?

No. Pattern-based detection, not a guarantee: it matches column names and value shapes such as email addresses, card numbers and dates. Names in free text, unusual formats, misspelled headers and data hidden in a single long text column can all slip through. A clean result means no sensitive pattern was found in the sampled rows — it does not prove the dataset is safe. Spot-check the data yourself before publishing or training on it.

Does my dataset upload anywhere?

No. Your CSV is parsed and scanned entirely in your browser with plain JavaScript — no libraries, no server, no account. The only third-party requests are the site's own analytics, which count page views. You can verify it yourself with the network-request counter in the privacy section below the checker. Do not paste real personal data into any website, this one included, without need — use a small extract for the check.

What counts as a sensitive column?

A column is marked Sensitive when it holds data that directly identifies a person or is a special category: emails, phone numbers, government ID numbers, card numbers, IBANs, IP addresses, dates of birth, names, street addresses, passwords or secrets, health data, financial data and identity document numbers. It is marked Possibly sensitive when it holds a quasi-identifier — a postcode, a city or coordinates, an age, a gender, a username, a date, a URL or a record id — which is low risk alone but can identify someone in combination.

How are names detected?

Names are detected from the column header only — first_name, last_name, full_name, customer_name and similar. Values alone cannot reliably identify a name, so a column called "col_7" full of names will not be flagged. Rename columns honestly before scanning, or check free-text columns by eye.

What is a quasi-identifier?

A field that does not identify anyone on its own but can single out a person when combined with others — the classic example is postcode plus age plus gender, which can narrow a person down to a handful of households. The checker warns you when two or more quasi-identifiers appear together and recommends generalising or dropping at least one of each pair.

What should I do with a sensitive column?

Every finding carries its own recommendation: remove government IDs, card numbers and secrets entirely; replace emails with a one-way hash if rows must stay joinable; turn dates of birth into age bands and postcodes into regions; replace record ids with random pseudonyms. The downloadable report lists the recommendation for every flagged column, so you can work through it methodically.

How many rows are scanned?

Up to 1,000 rows, sampled evenly across the file so a large dataset is checked along its whole length rather than just the top. Pasted text is capped at 200,000 characters and files at 5 MB, so a very large file should be scanned as a smaller extract first. A column is flagged only when at least 20% of its sampled values match a pattern, so one stray value does not condemn a column.

What file formats work?

CSV only — comma, semicolon, tab or pipe separated, with a header row. Spreadsheets and JSON are not supported: export to CSV first. The checker detects the delimiter from the first line and handles quoted fields, escaped quotes and CRLF line endings.