TOR, Signal, NYM, and NullWire solve different parts of the privacy problem. None of them solve all of it. This page shows what each protocol actually protects, where it fails, and what NullWire is building toward.
| Dimension | TOR | Signal | NYM | NullWire |
|---|---|---|---|---|
| Routing | Onion routing. 3-hop circuits. No mixing, no cover traffic. Vulnerable to end-to-end timing correlation. | Client-server. Messages go through Signal's centralized servers. No routing-layer privacy. | Loopix mixnet. Sphinx packets, Poisson delays, cover traffic. Cosmos-based coordination. | Loopix mixnet. Sphinx packets, 3-hop, X25519, fixed 2048-byte packets. Poisson delays + cover traffic. Solana coordination. LIVE |
| Metadata protection | Partial. Hides IP from destination, but exit nodes see traffic. Timing correlation defeats anonymity against state-level adversaries. | Weak. Server sees who talks to whom, when, how often. Phone number required. Metadata is the product. | Strong. Mixnet architecture makes timing correlation fundamentally harder. Cover traffic masks real traffic patterns. | Strong. Cover traffic + Poisson delays + fixed packet sizes eliminate timing and size correlation at every hop. LIVE |
| Timing analysis resistance | None. Well-documented vulnerability. State actors correlate entry/exit timing in minutes. | None. Centralized servers create obvious timing correlations. | Strong. Poisson-distributed inter-packet delays reduce correlation. | Strong. Three tunable delay profiles: Fast (~211ms), Balanced (~503ms), Private (~1.25s). Benchmarked on real hardware. LIVE |
| Message encryption | None at protocol level. Relies on HTTPS or application-layer encryption. Exit node sees plaintext if no TLS. | Excellent. Double Ratchet + X3DH. Gold standard for content encryption. Forward secrecy per message. | Sphinx in transit. Application-layer E2E depends on implementation. | X25519 Sphinx routing encryption. X3DH-style bootstrap + Double Ratchet-style E2E with key zeroization, session management, message lifecycle controls. Pre-alpha with significant hardening. Not yet externally audited. PRE-ALPHA E2E |
| Post-quantum crypto | Not resistant. Uses AES-256, RSA. No PQ roadmap published. | PQXDH deployed (ML-KEM + X25519 hybrid). Leading on this front. | Not resistant. Sphinx uses Curve25519. | ML-KEM-1024 + X25519 hybrid key exchange shipped (FIPS 203). HKDF-SHA3-256 key derivation (FIPS 202). SHIPPED |
| Decentralization | Volunteer-run nodes. No economic incentive. Some exit nodes are honeypots. Directory authorities are centralized. | Fully centralized. Signal Foundation runs all servers. Single point of legal seizure. | Cosmos blockchain. Token-incentivized nodes. Decentralized validator set. | Solana control plane. 2-of-3 validator-backed admission. Multi-RPC quorum for directory trust. No central kill switch. LIVE |
| Identity requirement | None. Connect via Tor Browser, no registration. | Phone number required. Linked to real identity in most jurisdictions. | zk-nym credentials (zero-knowledge). Fiat payment available. Node operators stake on-chain (linkable). End-user identity is credential-shielded. | None. No phone, no email, no persistent identifier. ZK identity planned. LIVE |
| Latency | Variable. 500ms–2s typical per circuit. Usable for web browsing. | Near-instant. Milliseconds. Centralized = fast. | Configurable. 1–10s typical for strong privacy. High latency is the trade-off. | Tunable. 211ms (fast) to 1.25s (max privacy). Benchmarked on devnet. LIVE |
| Maturity | 20+ years. Battle-tested. Millions of users. Known vulnerabilities are well-documented. | 10+ years. 100M+ installs. Widely trusted. Closed-source server. | NymVPN launched March 2025. 10K+ users. Multi-platform (iOS, Android, desktop). Actively developed. First real consumer mixnet product. | Pre-alpha. Devnet only. Proven cross-device. Not production-ready. PRE-ALPHA |
| Cost | Free. Volunteer infrastructure. | Free. Donation-funded. | Token-based. NYM token for bandwidth. Pay-per-use model. | Free during pre-alpha. Planned: $9–29/mo consumer tiers. PLANNED |
Message-layer E2E encryption is implemented with significant hardening. X3DH-style bootstrap with ML-KEM-1024 hybrid (FIPS 203) + Double Ratchet-style per-message forward secrecy with HKDF-SHA3-256 (FIPS 202), key zeroization on drop, skipped key expiry (48h), state file permissions (chmod 600), session reset with peer notification. Tested cross-device. Not yet subjected to third-party cryptographic review. NullWire now protects both who talks to whom and what they say — in pre-alpha with hardening.
Post-quantum key encapsulation is implemented. ML-KEM-1024 + X25519 hybrid key exchange is working and tests are passing. Signal has PQXDH deployed since 2023.
Production readiness does not exist. This is a devnet pre-alpha. Two-machine proof. Not a product. The protocol architecture is proven; the implementation is not hardened.
We state this because privacy tools that overclaim are more dangerous than no tool at all.
TOR uses onion routing: your message is encrypted in three layers and passed through three relays. Each relay strips one layer. This works for hiding your IP, but there is a fundamental problem: packet timing is preserved. An adversary watching both ends of the circuit can correlate entry and exit timing and deanonymize you in minutes.
A mixnet is different. Every node delays packets according to a Poisson distribution, reorders them, and injects cover traffic (fake messages that are indistinguishable from real ones). The result: even if you watch every single node in the network, you cannot correlate sender to receiver by timing alone.
NullWire adds another layer: fixed 2048-byte packets. Every message, every cover packet, every hop looks identical in size. There is no size-based correlation attack possible. TOR does not do this. Signal does not do this. This is the architectural trade-off NullWire makes: higher latency for fundamentally stronger anonymity guarantees.
This is not theory. NullWire's Loopix implementation has been benchmarked across three delay profiles on real separate-device hardware. Not localhost. Not simulation. Real Sphinx packets, real network hops, real measured latency.
| Delay Profile | Median Latency | Privacy Level |
|---|---|---|
| Fast | ~211ms | Baseline routing only. No timing obfuscation. |
| Balanced | ~503ms | Moderate. Usable for near-real-time messaging. |
| Private | ~1.25s | Full Poisson delays. Maximum timing resistance. |
Measured on devnet across two physical machines. Not localhost theatre.
← Back to nullwire.xyz