Flashnet: Protocol to Prototype
Mark Simkin, Mateusz Morusiewicz
Session summary
The presenters introduce Flashnet, an anonymous broadcast protocol that targets strong sender anonymity with low latency by minimizing network hops while accepting higher bandwidth costs. They explain that clients encode messages into randomly indexed arrays so that only the aggregate array is revealed, allowing messages to be recovered without linking them to specific senders. To keep communication practical, large arrays are encrypted under ephemeral keys, and only these small keys are secret-shared and added via a key-additive homomorphic encryption scheme. The authors stress that liveness relies on client-side trusted execution environments, but anonymity is preserved even if a TEE is broken.
To handle malicious servers without server-side TEEs, the design uses additive vector commitments so that server outputs can be checked against client commitments and disputes resolved. The implementation relies on lattice-based, post‑quantum primitives and the Chipmunk commitment system, which the authors report as not being a computational bottleneck. Prototype benchmarks with hundreds of clients and kilobyte-scale messages suggest that computation remains under a second, with bandwidth and link speed dominating performance and improved via aggregation and erasure coding on commodity networks.
Key takeaways
- The authors aim for strong anonymity against many corrupt parties and low latency by reducing network hops, trading off a moderate increase in bandwidth usage compared with other anonymous communication systems.
- Client-side trusted execution environments support liveness by enforcing well‑formed inputs, while anonymity is designed to hold even if a TEE is compromised, at worst causing denial-of-service rather than deanonymization.
- An implementation using post‑quantum key-additive homomorphic encryption, vector commitments, aggregation, and erasure coding shows that practical performance is primarily constrained by bandwidth, not computation, in the reported benchmarks.