Skip to content

Honest-But-Curious

An honest-but-curious party follows the protocol but tries to learn extra information from messages, metadata, or outputs.

Practical Example

A federated learning coordinator aggregates updates correctly but inspects individual updates for signals about a hospital's patient population.

Good Fit

This model is useful for early designs, mutually governed collaborations, and systems where protocol deviation is unlikely.

It is also useful as a narrow review step: first ask whether a party can learn too much while following the rules, then separately ask whether the rules are enforceable.

Limits

It does not cover malformed inputs, poisoning, selective aborts, query abuse, or compromised participants.

PET Implications

MPC, PSI, HE, secure aggregation, and clean rooms often start here. Add stronger controls when incentives are not aligned.

Review Questions

  • What messages, outputs, logs, and metadata does the curious party see?
  • Can the party repeat the protocol with slightly changed inputs or cohorts?
  • Does the claim still hold if the party combines protocol views with auxiliary data?
  • Which controls become mandatory if this party stops following the protocol?