Notesfoundations🌱 seedling

Data Minimization

Also: Data Minimisation · Collection Limitation

Data minimization means collecting and keeping only the personal data a specific purpose needs, in the least identifying form that serves it. Data you never stored can't be breached, subpoenaed, repurposed or sold.

Why it exists

Every other control (encryption, access control, deletion) can be misconfigured later. Absent data can't. Minimization also shrinks every downstream obligation: less to secure, delete and explain.

How it works

  • Derive, then discard: verify a passport, keep over_18 = true.
  • Coarsen: birth year not birth date, city not GPS track, timestamps rounded.
  • Aggregate early: daily counts instead of per-user event rows.
  • Separate identifiers: keep the real ID in one service, pseudonymous tokens elsewhere.
  • Don't log payloads: request logs are where personal data spreads unreviewed.
  • Beware that "anonymized" data is often re-identifiable by joining on quasi-identifiers: ZIP + birth date + sex was unique for about 87% of Americans in 1990 census data.

Trade-offs

Gains Costs
smaller breach impact some analysis becomes impossible
less to delete and secure harder debugging without payload logs
little to hand over under subpoena product teams lose "maybe later" data

In practice

  • Signal: its published responses to US subpoenas show it could provide only an account's creation date and last connection date, because nothing else is stored.
  • GDPR Article 5(1)(c): data must be "adequate, relevant and limited to what is necessary".

Pitfalls

  • "We might need it later" as a purpose: it isn't one.
  • Removing names and calling the result anonymous: remaining fields are join keys.

Recall

These cards are also in your review deck, scheduled with spaced repetition.