Privacy Attacks and Motivations for Formal Privacy
Stub. Why ad hoc privacy/anonymization fails in practice: historical motivation for information-theoretic secrecy (talking drums, the one-time pad), then concrete reconstruction and re-identification attacks — the Dinur-Nissim attack showing that answering too many statistical queries too accurately lets an attacker reconstruct an entire private database — plus attacks on security and on valuation/auction mechanisms.
Corresponds to CS/ECE 598DA, Weeks 1–2. Motivates the entire course: ad hoc approaches to secrecy and privacy (encoding messages so only an intended party understands them, or “anonymizing” a dataset by stripping identifiers) have a long history of looking secure and then failing. The one-time pad review (classical-cryptography-and-perfect-secrecy covers it formally) and talking drums (an early long-distance encoding scheme) are historical framing for this. The technical core is the Dinur-Nissim attack: an attacker who can ask enough statistical queries about a database, even queries individually “noised” or imprecise, can combine the answers to reconstruct the entire database almost exactly — the result that motivates differential privacy’s much stronger guarantee in differential-privacy-mechanisms-and-lower-bounds. Also covers analogous attacks on security mechanisms and on valuation/auction schemes that leak more than intended.
Outline (TODO — flesh out each)
- Motivations — why ad hoc privacy fails
- Historical framing: talking drums, the one-time pad (review)
- The Dinur-Nissim reconstruction attack
- Attacks on privacy
- Attacks on security
- Attacks on valuation / auction mechanisms