Privacy and de-identification: who removes personal data

In every program we cover, the buyer de-identifies the data, not you. Personal and identifying information (names, contact details, credentials, customer and client identities) is removed or replaced before the data reaches an AI lab. You decide what is in scope before anything moves, and the contract sets how the data is protected and when raw copies are deleted.

What each buyer says it does

Buyer	Who de-identifies	How they describe it
micro1	micro1	Personal and identifying information removed or transformed before data is shared with third parties; some datasets get synthetic rewrites
Mercor	Mercor	Automated extraction and anonymisation pipeline; no AI lab or vendor sees raw data
Nyne	Nyne	Names, identifiers and customer details removed before release; whole categories can be excluded
Troveo	Troveo	Names, contact details, credentials and client identities scrubbed to one documented standard, recorded in a manifest

These descriptions come from each buyer’s own pages. None of them publishes a technical standard you can audit in advance, so ask for one.

What you control

• Scope. Which systems, date ranges, teams and data types are included. Mercor says no data leaves your systems until you approve it.
• Exclusions. Whole categories you leave out: HR files, legal matters, health information, board materials.
• Samples. micro1 lets companies review representative samples before use. Ask every buyer for the same.
• Deletion. micro1 and Nyne describe retention and deletion terms agreed per deal. Get the schedule in writing.

Security points to check

micro1 says each company’s data is kept in isolated pipelines with access limited to authorised staff. Mercor lists encryption in transit, isolated credentials, SOC 2 Type II and keys managed in AWS KMS. Troveo needs no access to your systems because you export the data yourself.

Contract checklist

1. A written description of the de-identification method and what counts as personal information.
2. A list of excluded data categories.
3. Who may receive the data: micro1, for example, limits buyers to the US and allied countries.
4. Whether the data can be resold or sublicensed, and on what terms.
5. A retention period and deletion process for raw data, with confirmation when it is done.
6. What happens if personal information is found after delivery.

De-identification reduces privacy risk; it does not decide whether you may license the data at all. That depends on what you told employees and customers. See legal considerations.