Capability Manifold Surveillance: Topological Detection of Model Distillation via Activation Fingerprint Geometry
Abstract
Activation fingerprint geometry repurposed as a security primitive. Detects industrial-scale model distillation attacks by monitoring the model own activation space rather than inputs, bypassing the OMED impossibility theorem. AUC 1.000 on systematic attacks at scales 1.5B+. Continuous Proof-of-Humanity scoring achieves perfect separation.
Fields of Study & Subjects
Adversarial Robustness in Machine Learning
Security and Verification in Computing
Physical Unclonable Functions (PUFs) and Hardware Security
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 1 March 2026
- Language
- EN
- Document type
- Preprint