Introduction
Industrial Control Systems (ICS) form the backbone of modern society, managing everything from power grids to water treatment facilities. As these systems become increasingly digitized, they also become prime targets for sophisticated cyber-physical attacks. Regulators and Information Sharing and Analysis Centers (ISACs) are tasked with ensuring the resilience of these networks. However, they face a fundamental "regulatory blind spot": they cannot verify if a utility is correctly detecting and reporting attacks without accessing raw, sensitive sensor data—data that utilities are often legally or strategically prohibited from sharing.
The Privacy-Security Trade-off
The tension between operational privacy and regulatory oversight creates a significant security risk. If a utility’s IT or OT layers are compromised, an adversary could suppress legitimate attack alarms to maintain persistence. Conversely, a utility might be tempted to hide security lapses to avoid fines or reputational damage. To solve this, a system is needed that enables "computational transparency" rather than "data transparency"—allowing a utility to prove it is following security protocols without revealing the underlying proprietary data.
The zkSTAR Framework
The authors propose zkSTAR, a framework that wraps traditional state-space anomaly detection in a zero-knowledge (ZK) cryptographic envelope. At its core, zkSTAR uses an Extended Kalman Filter (EKF) to model the non-linear dynamics of an ICS. The EKF predicts expected behavior; any significant deviation (residual) between the prediction and actual sensor readings triggers an alarm.
To make this process verifiable, zkSTAR employs zk-SNARKs (specifically using the Halo2 backend). This allows the utility (the prover) to generate a mathematical proof that their EKF calculations and subsequent statistical tests were performed correctly according to a pre-agreed model.
The system is designed with a modular architecture to handle the computational weight of ZK circuits:
- Temporal Consistency (TC) Kernel: Ensures that state estimates are correctly propagated over time.
- Statistical Consistency (SC) Kernel: Verifies that the hypothesis tests (e.g., $\chi^2$ tests) used for alarm triggering are mathematically valid.
Verifying Complex Engineering Models
One of the primary challenges in applying ZK to engineering systems is the complexity of operations like Singular Value Decomposition (SVD) and non-linear solvers. These are computationally prohibitive to execute directly within a ZK circuit. zkSTAR overcomes this by using a "Compute-Offline, Validate-Online" strategy. Complex operations are performed outside the circuit, and the ZK proof only verifies that the results are consistent with the inputs through reconstruction checks.
Performance and Results
The framework was validated using the HAI (HIL-based Augmented ICS) and ORNL-PS (Oak Ridge National Laboratory Power System) datasets. The results demonstrate that zkSTAR maintains the high detection accuracy of standard EKFs while providing cryptographic guarantees.
Key performance metrics include:
- Verification Speed: Regulators can verify a proof in less than 1.5 seconds, making the system highly efficient for oversight.
- Prover Overhead: While generating proofs is resource-intensive (requiring significant RAM and CPU), it remains within the reach of modern utility-grade server hardware and is highly parallelizable.
- Integrity: The "knowledge-soundness" of zk-SNARKs ensures that a utility cannot suppress a valid alarm without failing the verification process.
Conclusion
zkSTAR represents a significant step forward in critical infrastructure security. By decoupling the ability to verify security compliance from the need to access raw data, it provides a path toward robust, privacy-preserving regulatory oversight. Future work will focus on optimizing the proving phase and extending the framework to even more complex industrial models.
Summary updated