Malicious Adversary

A malicious adversary may deviate from the protocol, submit malformed inputs, manipulate outputs, or selectively abort.

Practical Example

A participant in federated learning sends poisoned updates while secure aggregation makes individual update inspection difficult.

Use this model when parties have competing incentives, financial stakes, or weak governance.

Prefer malicious-secure protocols, input validation, robust aggregation, audit logs, rate limits, and incident response plans.

Many prototypes claim privacy against curious parties while product teams assume protection against malicious ones.

Design area	Honest-but-curious design	Malicious-adversary review
Inputs	Well-formed inputs are assumed	Validate format, range, provenance, and replay behavior
Participants	Identity is mostly administrative	Authenticate, rate-limit, suspend, and audit participants
Computation	Protocol privacy is the main concern	Add robustness, malicious security, or verifiable execution
Failure handling	Retry failed runs	Treat aborts, dropouts, and timing as possible signals
Operations	Debugging can inspect more artifacts	Debugging paths must not bypass privacy controls