Keyvan Hardani
Biography
Keyvan Hardani is an AI & Agentic Systems Lead Engineer and researcher based in Munich, Germany, specializing in AI Safety, Adversarial Machine Learning, and Automotive Cybersecurity. Currently working as Project Engineer for Automotive System and Integration Testing at MAN Truck & Bus SE, he leads HiL Security teams and develops AI-driven security solutions including Reinforcement Learning Firewalls and CNN-based Intrusion Detection Systems. His research focuses on adversarial trigger generation, reward model security, and real-time automotive network protection. He completed his Master of Science in Cyber Security at Hochschule der Bayerischen Wirtschaft (HDBW) Munich, with his thesis on manipulation of neural language models and defense strategies against trojaned AI systems.
Email: keyvan.hardani@ieee.org
Location: Munich, Germany
ORCID: 0009-0000-6003-8826
Education
-
M.Sc. in Cyber Security
Hochschule der Bayerischen Wirtschaft (HDBW), Munich (2021-2025)
Master's Thesis: Manipulation of Neural Language Models - Security Risks and Defense Strategies for Trojaned AI Models -
B.Sc. in Computer Engineering
Institute of Engineering & Management (2008-2012)
Research Interests
- AI Safety & Robustness
- Adversarial Machine Learning & Backdoor Detection
- Reward Model Security (RLHF & LLM Safety)
- Automotive Cybersecurity & ISO 21434 Compliance
- Intrusion Detection Systems (IDS) for CAN/LIN Networks
- Neuromorphic Computing & Real-time AI
- Embedded Systems Security
Professional Experience
-
Project Engineer - Automotive HiL & AI Security (Team Lead)
MAN Truck & Bus SE (via Vdynamics GmbH), Munich (May 2024 - Present) -
Senior Full-Stack & AI Developer
Kapital Medien GmbH & SDK e.V., Munich (Apr 2021 - 2023)
Professional Affiliations
- IEEE Member - Institute of Electrical and Electronics Engineers
- Senior Researcher, HDBW University of Applied Sciences, Munich
- Vulnerability Researcher, Wordfence Security
Certifications
- Offensive Security Certified Professional (OSCP)
- Certified Ethical Hacker 5.0 (CEH)
- Automotive Cybersecurity Training - ISO 21434 (Vector Informatik)
- ISTQB Certified Tester Foundation Level
- Harvard University CS50 - Artificial Intelligence
- MATLAB & Simulink Certification
- Foundations of Cybersecurity (Google)
News
- [2025.01] CVE-2025-0990 (CSRF vulnerability) published - CVSS 4.3
- [2024.11] SecIDS-CNN model published on Hugging Face with 97.72% accuracy
- [2024.10] Master Thesis completed on trojaned AI models and defense strategies
- [2024.09] Joined IEEE focusing on AI Safety and Cybersecurity standards
- [2024.05] Started Team Lead position at MAN Truck & Bus for HiL Security
- [2023.12] AI-Driven SIEM system published with real-time threat detection
Selected Publications | [More in Google Scholar]
-
Manipulation of Neural Language Models: Security Risks and Defense Strategies for Trojaned AI Models
Keyvan Hardani
Master's Thesis, HDBW Munich, 2024 -
Verbesserung von Belohnungsmodellen durch Adversariale Trigger-Generierung
[PDF]
Keyvan Hardani and Collaborators
Technical Report on AI Safety and Cybersecurity, 2024 -
SecIDS-CNN: Advanced Convolutional Neural Network for Intrusion Detection
[Model]
[Code]
Keyvan Hardani
Hugging Face & GitHub, 2024
High-performance CNN model achieving 97.72% accuracy for automotive IDS
AI Models & Research Projects
- SecIDS-CNN - CNN-based Intrusion Detection System (97.72% accuracy) [HuggingFace] [GitHub]
- CANDefender-Fuzzy - LSTM-based Fuzzy Attack Detection for CAN Bus (94.09% accuracy) [HuggingFace]
- CANDefender-DoS - LSTM-based DoS Attack Detection for CAN Bus (94.06% accuracy) [HuggingFace]
- AnomalyDetection-MVTech-Metal - Deep Learning for Industrial Quality Control [HuggingFace]
- Vulna v4.0 - AI-Powered Penetration Testing Framework [GitHub]
Security Achievements (CVEs)
- CVE-2025-0990 - Cross-Site Request Forgery (CSRF) - CVSS 4.3 [NVD]
- CVE-2021-4455 - Remote Code Execution (RCE) - CVSS 9.8 [NVD]
- CVE-2021-24997 - Information Disclosure - CVSS 6.5 [NVD]
Selected Open-Source Projects
AI-Driven SIEM with Real-time Threat Detection
Comprehensive Security Information and Event Management system powered by CNNs and NLP. Provides intelligent threat detection, automated response mechanisms, and real-time security monitoring for enterprise environments with Groq integration.
Python, TensorFlow, CNNs, NLP, SIEM, Groq
Autonomous XGO Robot Navigation System
Advanced autonomous navigation system for quadruped robotic platforms implementing computer vision, object detection, and real-time decision-making algorithms for environment interaction.
Python, OpenCV, Computer Vision, Robotics, AI
Acoustic Echo Cancellation Library
Implementation of adaptive filter algorithms including LMS, NLMS, and RLS for acoustic echo cancellation and audio signal processing applications.
Python, Signal Processing, Adaptive Filters, DSP
XGO2 Robot Expression System
Open-source implementation for animating robotic facial features with synchronized expressions. Enables lifelike eye and mouth movements coordinated with speech output for enhanced human-robot interaction.
Python, Robotics, HRI, Animation
WordPress-Streamlit Authentication Integration
Secure authentication framework integrating WordPress user management with Streamlit applications using JWT-based authentication for seamless single sign-on across platforms.
Python, PHP, JWT, WordPress, Streamlit
More projects available on GitHub.
Academic Service
Vulnerability Research:
- Wordfence Intelligence - WordPress Security Research & CVE Discovery
- Open-source security contributions and responsible disclosure
Research Collaboration:
- Collaborating with Prof. Jianmin Chen & Prof. Max Moser on AI Safety research
- Technical workshops on Automotive Cybersecurity and AI Safety