Data types
How do I de-identify expert annotations and labels for AI training?
By SourceX Editorial · Updated
Short answer
To de-identify expert annotations and labels, remove or replace annotator names, underlying personal data in labeled items and client identifiers, check free text and attachments for hidden details, review samples by hand and exclude anything doubtful. Your company then approves the prepared copy before it leaves.
What expert annotations and labels usually contain#
Typically labels, ratings, corrections and reviewer notes added by skilled staff. The work itself is what buyers value; the identities are not.
What to remove or replace#
At minimum: annotator names, underlying personal data in labeled items and client identifiers. Names are often swapped for consistent placeholders so conversations still make sense.
How to check it worked#
Automated tools catch most details, but not all. Reviewers spot-check samples, look for rare facts that could point to a person, and drop records they are unsure about. Buyers are also contractually barred from trying to identify anyone.
How SourceX handles it#
SourceX starts with a short fit check that shares no data. If it fits, rights are reviewed, a copy is prepared with personal and confidential details removed, and your company approves exactly what leaves before anything is delivered under a signed license.
Frequently asked questions
Can I do the de-identification myself?
You can, but SourceX or an approved preparation partner normally does it and you review the result.
Is this legal advice?
No. It is general information; your counsel should review any specific deal.
Related resources
- IndustryBPO & contact centers data
- IndustryRecruiting & staffing data
- QuestionShould companies sell or license their data?
- QuestionCan CRM data be licensed?
- InsightIs it safe to license company data for AI training?
- InsightHow is company data anonymized before AI training?
- InsightHow to remove sensitive details from business data for AI training
- SolutionData partnerships between businesses and AI developers
- SolutionWhat is AI evaluation data?
- DataCode review records
- DataCompany email archives
- GlossaryData processing agreement
See if your company qualifies
A short company assessment. No data uploads are needed.