← All posts
Text similarity to the original after de-identification (BERTScore)SurrogateShield preprint, arXiv 2605.13538 (2026)
Redaction vs replacement: why [PERSON] tags destroy data value
Replace every name in a conversation with [PERSON] and you've protected privacy, but you've also erased who said what to whom.
What consistent replacement means
- The same person always becomes the same stand-in: 'Person A'.
- Stand-ins match type: a company becomes a company, a date stays a date.
- The conversation still makes sense to a model.
The caveats
Research also shows pseudonymisation can reduce utility for some tasks, and variety in substitutes matters. Testing on a sample is essential.
Key takeaways
- Placeholders protect privacy but hurt utility.
- Consistent, typed surrogates keep more meaning.
- Always test re-identification risk afterwards.
Sources
What could your company's records earn?
A free, nine-question estimate. No files, no logins.
More in Privacy & law
Data brokers and national security: what the UK government is asking
DSIT's 2025 call for evidence asked how brokers keep data away from hostile actors. What it covered, and the checks every broker should run now.
Personal data that must never leave: a checklist
Special category data, criminal records, children's data and more: what to exclude before any data deal.
The motivated intruder test, step by step
The ICO's benchmark for whether de-identified data is safe: a determined, competent person with no special access.