The MyShape Protocol faces three classes of adversaries. Eight attack signatures are detected across four PES dimensions. Defense-in-depth across five layers.
Frequency entropy collapse + missing cross-joint jerk correlation
Uniform inter-frame timing + stale MV_hash + over-clean noise profile
Inter-frame consistency + frequency monitoring + multi-dimension corroboration
| Tier | Cost | Attack Types | Max Success |
|---|---|---|---|
| C0 — Low | < $1K | Replay, basic imitation | ~0% |
| C1 — Medium | $1K–$100K | Mocap, diffusion models | < 1% |
| C2 — High | $100K–$10M | Digital twin, adversarial training | < 5% |
| C3 — Extreme | > $10M | Full biological simulation | < 10% |
MyShape security property: the cost to forge presence exceeds the value of forging it.
Sybil resistance via PES uniqueness. One presence → one identity.
ZK-Presence: PoP + MP + EP. Verifiable without exposing motion data.
PES 4-dimensional entropy. Anti-replay. Anti-synthesis. Cross-joint correlation.
SST 18-point normalization. Device-agnostic. Manifold projection — non-invertible.
Raw sensor input processed on-device. Nothing stored. Nothing uploaded.
Pr[AI generates PES ≥ 0.65] → 0
The gap between biological entropy and AI-generated smoothness is a fundamental limit, not a technological one. It arises from the informational asymmetry between a living nervous system and any external model of it.