IEEE GLOBECOM 2026 · Quantum-Classical Cybersecurity

VQC-ZTI Variational Quantum Control for Zero Trust Protection of the Tactile Internet

A split-plane zero-trust framework that performs hybrid variational quantum anomaly scoring on mirrored encrypted-flow telemetry while keeping deterministic policy enforcement outside the quantum inference path.

Framework Overview

VQC-ZTI separates asynchronous anomaly-evidence generation from latency-critical zero-trust enforcement so that quantum inference does not become a synchronous hop in the current control path.

Hybrid Quantum-Classical Learning

A classical embedder, 12-qubit variational quantum circuit, and classical classification head are trained end-to-end through PyTorch and PennyLane.

🛡

Zero-Trust Enforcement

Risk evidence is separated from policy execution. Cached PEP actions provide deterministic grant, restrict, step-up, or deny enforcement.

Distribution-Shift Evaluation

The model is evaluated under random, entity-group, and temporal holdouts using CESNET-derived aggregated encrypted-flow telemetry.

4,875
Traffic Records
12
Input Features / Qubits
5
Independent Seeds
3
Evaluation Protocols

Split-Plane Architecture

Security evidence is generated asynchronously, while the immediate transaction remains on a lightweight deterministic enforcement path.

Off-Path Evidence Plane
Mirrored Encrypted-Flow Telemetry
Feature Preparation
Classical Embedder
Quantum Encoding
12-Qubit VQC / QNN
Classical Classification Head
Risk Fusion + Policy Computation
Versioned
Policy Updates
On-Path Enforcement Plane
Client / Tactile Request
Policy Enforcement Point
Cached Deterministic Policy Lookup
Grant · Restrict · Step-Up · Deny
Protected Tactile Internet Service
Key design principle: VQC inference is not an intermediate hop between the current request and the protected service. Delayed evidence generation can postpone a future policy update without directly delaying the current enforcement action.

Evaluation Results

Mean ± standard deviation across five runs for the full-hybrid QNN.

Evaluation Split ROC-AUC Accuracy FPR FPR Reduction vs. ExtraTrees
Random 0.9981 ± 0.0004 0.9756 ± 0.0042 0.0270 ± 0.0050 44.6%
Entity-Group 0.9974 ± 0.0009 0.9762 ± 0.0028 0.0241 ± 0.0042 49.6%
Temporal 0.9941 ± 0.0018 0.9747 ± 0.0044 0.0248 ± 0.0052 67.9%
The reported labels are statistical anomaly pseudo-labels rather than verified attack annotations. These experiments therefore do not establish verified intrusion detection or quantum computational advantage.

Experimental Configuration

Quantum Model

12 qubits · 2 variational layers · Pauli-Y angle encoding · nearest-neighbor CNOT entanglement · Pauli-Z expectation measurements.

Training

20 epochs · batch size 32 · learning rate 2×10⁻³ · random seeds 42–46 · class-weighted negative log-likelihood.

Execution Setting

Noiseless analytic expectation values. Finite-shot uncertainty, physical hardware noise, transpilation, and queueing effects are outside the reported evaluation.

Authors

Mubassir Serneabat Sudipto
Electrical and Computer Engineering
Iowa State University
Ames, Iowa, USA

msudipto@iastate.edu
Shakil Ahmed
Computer Science, College of Computing
Grand Valley State University
Allendale, Michigan, USA

ahmeshak@gvsu.edu
Ashfaq Khokhar
Carl R. Ice College of Engineering
Kansas State University
Manhattan, Kansas, USA

akhokhar@k-state.edu

Citation

If you use VQC-ZTI, the implementation, methodology, or reported experimental results, please cite the associated paper.

BibTeX
@misc{sudipto2026vqczti,
  title         = {{VQC-ZTI}: Variational Quantum Control for Zero Trust Protection of the Tactile Internet},
  author        = {Sudipto, Mubassir Serneabat and Ahmed, Shakil and Khokhar, Ashfaq},
  year          = {2026},
  eprint        = {2608.18572},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CR},
  doi           = {10.48550/arXiv.2608.18572},
  url           = {https://arxiv.org/abs/2608.18572},
  note          = {Accepted at IEEE Global Communications Conference (GLOBECOM 2026)}
}