MPC Threat Model and Security Definitions
What you'll learn: How security experts categorize attackers and define what "secure" means in multi-party computation.
Understanding MPC Adversaries
When multiple parties compute together without revealing their private inputs (as introduced in the previous lesson), we need to consider: who might try to cheat, and how?
MPC protocols are designed against two main adversary types:
Semi-Honest (Honest-but-Curious) Adversaries
These attackers follow the protocol rules but try to learn extra information by analyzing messages they receive. Think of a curious colleague who completes their work correctly but reads every email carefully hoping to discover secrets.
Example: A party correctly computes their share but records all intermediate values trying to reconstruct others' inputs.
Malicious Adversaries
These attackers can deviate arbitrarily from the protocol—sending fake values, aborting at strategic times, or colluding with others. This is like a participant who lies about their input or sabotages the computation.
Corruption Thresholds
Not all parties may be honest. The corruption threshold defines how many parties an adversary can control while the protocol remains secure:
- t < n/2: Honest majority (most common)
- t < n: Dishonest majority (stronger but less efficient)
With 5 parties and threshold t=2, the protocol tolerates up to 2 corrupted participants.
Security Guarantees
MPC protocols typically guarantee:
- Privacy: Corrupted parties learn nothing beyond the final output
- Correctness: The result matches what an honest computation would produce
- Independence: Inputs cannot be chosen based on others' values
Key Takeaway: MPC security depends on clearly defining adversary capabilities (semi-honest vs malicious), how many parties can be corrupted (threshold), and what guarantees the protocol provides (privacy, correctness).