Security Research · Dossier SRM Inst. of Science & Technology · 2026

Defenses that hold when
the model is fooled.

I'm Aarnav Singh — a final-year B.Tech CSE (Cybersecurity) student researching how Zero-Trust contextual policy can absorb adversarial machine-learning attacks, and building the security tooling around that research.

01
About

Security that assumes the model will be wrong.

I got into security through an internship at HCLTech, working across enterprise network security, Microsoft Azure infrastructure, and DevSecOps automation in Python. That raised a question that's stuck with me: as detection leans harder on machine learning, what happens when an attacker learns to fool the model itself?

My major project answers that for Zero-Trust network architectures. I built a six-layer system pairing an ML intrusion-detection classifier with a context-aware policy engine, then attacked it with domain-constrained FGSM and PGD methods — and measured how much Zero-Trust context could compensate, with no model retraining at all.

Alongside the research I built an applied counterpart — a full-stack Hybrid Behavioral Intrusion Detection System with a real-time dashboard, closer to a production SOC tool.

DegreeB.Tech CSE · Cybersecurity
InstituteSRM Inst. Science & Tech
Graduating2026
InternshipHCLTech · 2024
PaperUnder review · JISA
FocusZero-Trust · Adversarial ML
02
Research highlight

Zero-Trust context absorbs adversarial evasion.

● Under review · JISA, Elsevier · Scopus Q1 · IF 3.7

Adversarial Attack Detection in Zero-Trust Networks

A six-layer Python/PyTorch system showing that Zero-Trust contextual policy can absorb the damage of successful FGSM/PGD evasion against an ML intrusion-detection classifier — without retraining.

ML classifier aloneunder adversarial evasion
65%
+ Zero-Trust contextdevice · geo · identity · time
2.2%
↓ 96.7% fewer bypasses · zero retraining
90.7%
UNSW-NB15 accuracy
99.9%
CICIDS-2017 accuracy
5 seeds
averaged for robustness
FGSM / PGD
domain-constrained
03
Selected work

Built, attacked, and measured.

i

Adversarial Attack Detection in Zero-Trust Networks

Jan 2026 · Research system

A six-layer Zero-Trust security system combining an ML intrusion-detection classifier with a context-aware policy engine, stress-tested against domain-constrained adversarial attacks.

  • Trained classifiers on UNSW-NB15 and CICIDS-2017 — 90.7% and 99.9% accuracy across 5 seeds.
  • Implemented domain-constrained FGSM/PGD evasion attacks against the trained classifiers.
  • Reduced Zero-Trust bypass rates from 65% to 2.2% using contextual signals — with zero retraining.
  • Co-authored a resulting paper, under review at JISA (Elsevier, Scopus Q1, IF 3.7).
PythonPyTorchAdversarial MLZero-TrustUNSW-NB15CICIDS-2017
View repository ↗
ii

Hybrid Behavioral Intrusion Detection System

Feb 2026 · Full-stack tool

A full-stack IDS combining rule-based detection with Isolation-Forest anomaly detection, paired with a real-time monitoring dashboard.

  • Built a real-time React dashboard over WebSockets and REST APIs for live threat visualization.
  • Integrated Elasticsearch for centralized log indexing and threat-event correlation.
  • Deployed Prometheus to monitor detection latency and system performance.
  • Containerized the full system with Docker to simulate a scalable, Zero-Trust-aligned deployment.
ReactIsolation ForestElasticsearchPrometheusDockerWebSockets
View repository ↗
04
Experience

Where it started.

May 2024 — Jul 2024Noida, India

HCLTech — Intern

Network & Security
  • Built foundations in enterprise network and perimeter security — firewalls, segmentation, and threat-mitigation strategy.
  • Gained hands-on understanding of Microsoft Azure infrastructure while supporting security operations.
  • Applied Python for DevSecOps automation — security-tooling configuration and codebase review.
05
Capabilities

Toolkit.

Security & Monitoring

IDS/IPSSIEMSOARLog AnalysisIncident ResponseThreat IntelMITRE ATT&CKVuln Assessment

Machine Learning

Adversarial MLAnomaly DetectionIsolation ForestFeature EngineeringROC CurveConfusion Matrix

Networking & Zero-Trust

TCP/IPDNSVPNSegmentationFirewallsZTNASASESD-WANNAC

Tools & Platforms

ElasticsearchPrometheusWiresharkNmapNessusMetasploitFortiGateKaliWAF

Cloud & Development

AzureAWSIAMDevSecOpsLinuxPythonPyTorchDockerReact.jsREST

Certifications

ISC2 CCFortinet FCAFortinet FCFNPTEL ACNDeloitte (Forage)