The course aims to provide a solid understanding of the intersections between cybersecurity and artificial intelligence at various levels. Students will acquire skills in the typical uses of AI tools for mitigating cybersecurity risks, such as AI-driven attack detection and prevention.
The course will also cover new risks introduced by the use of AI in enterprise systems and business processes, along with mitigation techniques. Offensive techniques leveraging AI will be explored, including its use in malware, social engineering, phishing, attacks on CAPTCHA and biometric systems, and information manipulation.
Ethical issues related to the use of AI in cybersecurity systems will also be briefly discussed.
The course will also cover new risks introduced by the use of AI in enterprise systems and business processes, along with mitigation techniques. Offensive techniques leveraging AI will be explored, including its use in malware, social engineering, phishing, attacks on CAPTCHA and biometric systems, and information manipulation.
Ethical issues related to the use of AI in cybersecurity systems will also be briefly discussed.
Curriculum
teacher profile teaching materials
Overview of the various contexts in which artificial intelligence can be used within an organization, including benefits and threats
Defensive use of artificial intelligence
Data formats: host-based and network-based
Feature engineering
Anomaly detection
Malware analysis
Network traffic analysis: data preparation, supervised learning, unsupervised learning, ensemble learning
Adversarial machine learning
AI-related threats and mitigation strategies
Model poisoning and tampering, defenses
Supply chain attacks and defenses
Evasion attacks and defenses
Privacy attacks and defenses
Generative AI: GANs for malicious applications (deepfakes, voice cloning, malware evasion, spam, video surveillance), deepfake detection
LLMs: architectures leveraging LLMs, RAG, MCP, chain-of-thought, threat models
Direct/indirect prompt injection, jailbreaks, phishing, style injection, role-playing, impersonation, ChatGPT_DAN, data exfiltration, privilege escalation, and remote code execution (e.g., through MCP), defenses and mitigations
Poisoning in RAG systems and fine-tuning. Supply chain risks: publishing a poisoned LLM. Benchmarks, red teaming, monitoring, and incident response
Secure-by-design methodologies, frameworks, MLSecOps, and shift-left approaches
Programme
Fundamentals of cybersecurity and machine learningOverview of the various contexts in which artificial intelligence can be used within an organization, including benefits and threats
Defensive use of artificial intelligence
Data formats: host-based and network-based
Feature engineering
Anomaly detection
Malware analysis
Network traffic analysis: data preparation, supervised learning, unsupervised learning, ensemble learning
Adversarial machine learning
AI-related threats and mitigation strategies
Model poisoning and tampering, defenses
Supply chain attacks and defenses
Evasion attacks and defenses
Privacy attacks and defenses
Generative AI: GANs for malicious applications (deepfakes, voice cloning, malware evasion, spam, video surveillance), deepfake detection
LLMs: architectures leveraging LLMs, RAG, MCP, chain-of-thought, threat models
Direct/indirect prompt injection, jailbreaks, phishing, style injection, role-playing, impersonation, ChatGPT_DAN, data exfiltration, privilege escalation, and remote code execution (e.g., through MCP), defenses and mitigations
Poisoning in RAG systems and fine-tuning. Supply chain risks: publishing a poisoned LLM. Benchmarks, red teaming, monitoring, and incident response
Secure-by-design methodologies, frameworks, MLSecOps, and shift-left approaches
Attendance
-Type of evaluation
Written exam, oral exam, project teacher profile teaching materials
Overview of the various contexts in which AI can be used within an organization, including benefits and threats.
# Defensive use of AI
Data formats: host-based and network-based
Feature engineering
Anomaly detection
Malware analysis
Network traffic analysis: data preparation, supervised learning, unsupervised learning, ensembles
Adversarial ML
# AI-related threats and mitigation strategies
Model poisoning and tampering, defenses
Supply chain attacks and defenses
Evasion attacks and defenses
Privacy attacks and defenses
Generative AI: GANs for malicious applications such as deepfakes, voice cloning, evasion, malware, spam, video surveillance; deepfake detection
LLMs: architectures leveraging LLMs, RAG, MCP, chain-of-thoughts, threat models
Direct and indirect prompt injection, jailbreaks, phishing, style injection, role-playing, impersonation, ChatGPT DAN, data exfiltration, privilege escalation and remote code execution, for example through MCP; defenses and mitigations
Poisoning in RAG and fine-tuning. Supply chain risks: publishing a poisoned LLM. Benchmarks. Red teaming. Monitoring, incident response.
Secure-by-design methodology, frameworks, MLSecOps, shift-left.
https://www.proquest.com/docview/2134682020/73CEC526C564F85PQ
- Adversarial AI Attacks, Mitigations, and Defense Strategies: A cybersecurity professional's guide to AI attacks, threat modeling, and securing AI with MLSecOps. John Sotiropoulos Sotiropoulos
https://www.proquest.com/docview/3081482260/3B9D9980BFF8481BPQ
Programme
Cybersecurity and machine learning reviewOverview of the various contexts in which AI can be used within an organization, including benefits and threats.
# Defensive use of AI
Data formats: host-based and network-based
Feature engineering
Anomaly detection
Malware analysis
Network traffic analysis: data preparation, supervised learning, unsupervised learning, ensembles
Adversarial ML
# AI-related threats and mitigation strategies
Model poisoning and tampering, defenses
Supply chain attacks and defenses
Evasion attacks and defenses
Privacy attacks and defenses
Generative AI: GANs for malicious applications such as deepfakes, voice cloning, evasion, malware, spam, video surveillance; deepfake detection
LLMs: architectures leveraging LLMs, RAG, MCP, chain-of-thoughts, threat models
Direct and indirect prompt injection, jailbreaks, phishing, style injection, role-playing, impersonation, ChatGPT DAN, data exfiltration, privilege escalation and remote code execution, for example through MCP; defenses and mitigations
Poisoning in RAG and fine-tuning. Supply chain risks: publishing a poisoned LLM. Benchmarks. Red teaming. Monitoring, incident response.
Secure-by-design methodology, frameworks, MLSecOps, shift-left.
Core Documentation
- Machine Learning and Security : Protecting Systems with Data and Algorithms. Chio, Clarence; Freeman, Davidhttps://www.proquest.com/docview/2134682020/73CEC526C564F85PQ
- Adversarial AI Attacks, Mitigations, and Defense Strategies: A cybersecurity professional's guide to AI attacks, threat modeling, and securing AI with MLSecOps. John Sotiropoulos Sotiropoulos
https://www.proquest.com/docview/3081482260/3B9D9980BFF8481BPQ
Attendance
Students are encouraged to attend to classes.Type of evaluation
Project and discussion on the first part. Oral exam, or written exam depending on enrollment numbers, on the second part. teacher profile teaching materials
Overview of the various contexts in which artificial intelligence can be used within an organization, including benefits and threats
Defensive use of artificial intelligence
Data formats: host-based and network-based
Feature engineering
Anomaly detection
Malware analysis
Network traffic analysis: data preparation, supervised learning, unsupervised learning, ensemble learning
Adversarial machine learning
AI-related threats and mitigation strategies
Model poisoning and tampering, defenses
Supply chain attacks and defenses
Evasion attacks and defenses
Privacy attacks and defenses
Generative AI: GANs for malicious applications (deepfakes, voice cloning, malware evasion, spam, video surveillance), deepfake detection
LLMs: architectures leveraging LLMs, RAG, MCP, chain-of-thought, threat models
Direct/indirect prompt injection, jailbreaks, phishing, style injection, role-playing, impersonation, ChatGPT_DAN, data exfiltration, privilege escalation, and remote code execution (e.g., through MCP), defenses and mitigations
Poisoning in RAG systems and fine-tuning. Supply chain risks: publishing a poisoned LLM. Benchmarks, red teaming, monitoring, and incident response
Secure-by-design methodologies, frameworks, MLSecOps, and shift-left approaches
Mutuazione: 20830030 Cybersecurity and Artificial Intelligence in Ingegneria informatica e dell'intelligenza artificiale LM-32 IANNUCCI STEFANO, PIZZONIA MAURIZIO
Programme
Fundamentals of cybersecurity and machine learningOverview of the various contexts in which artificial intelligence can be used within an organization, including benefits and threats
Defensive use of artificial intelligence
Data formats: host-based and network-based
Feature engineering
Anomaly detection
Malware analysis
Network traffic analysis: data preparation, supervised learning, unsupervised learning, ensemble learning
Adversarial machine learning
AI-related threats and mitigation strategies
Model poisoning and tampering, defenses
Supply chain attacks and defenses
Evasion attacks and defenses
Privacy attacks and defenses
Generative AI: GANs for malicious applications (deepfakes, voice cloning, malware evasion, spam, video surveillance), deepfake detection
LLMs: architectures leveraging LLMs, RAG, MCP, chain-of-thought, threat models
Direct/indirect prompt injection, jailbreaks, phishing, style injection, role-playing, impersonation, ChatGPT_DAN, data exfiltration, privilege escalation, and remote code execution (e.g., through MCP), defenses and mitigations
Poisoning in RAG systems and fine-tuning. Supply chain risks: publishing a poisoned LLM. Benchmarks, red teaming, monitoring, and incident response
Secure-by-design methodologies, frameworks, MLSecOps, and shift-left approaches
Attendance
-Type of evaluation
Written exam, oral exam, project teacher profile teaching materials
Overview of the various contexts in which AI can be used within an organization, including benefits and threats.
# Defensive use of AI
Data formats: host-based and network-based
Feature engineering
Anomaly detection
Malware analysis
Network traffic analysis: data preparation, supervised learning, unsupervised learning, ensembles
Adversarial ML
# AI-related threats and mitigation strategies
Model poisoning and tampering, defenses
Supply chain attacks and defenses
Evasion attacks and defenses
Privacy attacks and defenses
Generative AI: GANs for malicious applications such as deepfakes, voice cloning, evasion, malware, spam, video surveillance; deepfake detection
LLMs: architectures leveraging LLMs, RAG, MCP, chain-of-thoughts, threat models
Direct and indirect prompt injection, jailbreaks, phishing, style injection, role-playing, impersonation, ChatGPT DAN, data exfiltration, privilege escalation and remote code execution, for example through MCP; defenses and mitigations
Poisoning in RAG and fine-tuning. Supply chain risks: publishing a poisoned LLM. Benchmarks. Red teaming. Monitoring, incident response.
Secure-by-design methodology, frameworks, MLSecOps, shift-left.
https://www.proquest.com/docview/2134682020/73CEC526C564F85PQ
- Adversarial AI Attacks, Mitigations, and Defense Strategies: A cybersecurity professional's guide to AI attacks, threat modeling, and securing AI with MLSecOps. John Sotiropoulos Sotiropoulos
https://www.proquest.com/docview/3081482260/3B9D9980BFF8481BPQ
Mutuazione: 20830030 Cybersecurity and Artificial Intelligence in Ingegneria informatica e dell'intelligenza artificiale LM-32 IANNUCCI STEFANO, PIZZONIA MAURIZIO
Programme
Cybersecurity and machine learning reviewOverview of the various contexts in which AI can be used within an organization, including benefits and threats.
# Defensive use of AI
Data formats: host-based and network-based
Feature engineering
Anomaly detection
Malware analysis
Network traffic analysis: data preparation, supervised learning, unsupervised learning, ensembles
Adversarial ML
# AI-related threats and mitigation strategies
Model poisoning and tampering, defenses
Supply chain attacks and defenses
Evasion attacks and defenses
Privacy attacks and defenses
Generative AI: GANs for malicious applications such as deepfakes, voice cloning, evasion, malware, spam, video surveillance; deepfake detection
LLMs: architectures leveraging LLMs, RAG, MCP, chain-of-thoughts, threat models
Direct and indirect prompt injection, jailbreaks, phishing, style injection, role-playing, impersonation, ChatGPT DAN, data exfiltration, privilege escalation and remote code execution, for example through MCP; defenses and mitigations
Poisoning in RAG and fine-tuning. Supply chain risks: publishing a poisoned LLM. Benchmarks. Red teaming. Monitoring, incident response.
Secure-by-design methodology, frameworks, MLSecOps, shift-left.
Core Documentation
- Machine Learning and Security : Protecting Systems with Data and Algorithms. Chio, Clarence; Freeman, Davidhttps://www.proquest.com/docview/2134682020/73CEC526C564F85PQ
- Adversarial AI Attacks, Mitigations, and Defense Strategies: A cybersecurity professional's guide to AI attacks, threat modeling, and securing AI with MLSecOps. John Sotiropoulos Sotiropoulos
https://www.proquest.com/docview/3081482260/3B9D9980BFF8481BPQ
Attendance
Students are encouraged to attend to classes.Type of evaluation
Project and discussion on the first part. Oral exam, or written exam depending on enrollment numbers, on the second part. teacher profile teaching materials
Overview of the various contexts in which artificial intelligence can be used within an organization, including benefits and threats
Defensive use of artificial intelligence
Data formats: host-based and network-based
Feature engineering
Anomaly detection
Malware analysis
Network traffic analysis: data preparation, supervised learning, unsupervised learning, ensemble learning
Adversarial machine learning
AI-related threats and mitigation strategies
Model poisoning and tampering, defenses
Supply chain attacks and defenses
Evasion attacks and defenses
Privacy attacks and defenses
Generative AI: GANs for malicious applications (deepfakes, voice cloning, malware evasion, spam, video surveillance), deepfake detection
LLMs: architectures leveraging LLMs, RAG, MCP, chain-of-thought, threat models
Direct/indirect prompt injection, jailbreaks, phishing, style injection, role-playing, impersonation, ChatGPT_DAN, data exfiltration, privilege escalation, and remote code execution (e.g., through MCP), defenses and mitigations
Poisoning in RAG systems and fine-tuning. Supply chain risks: publishing a poisoned LLM. Benchmarks, red teaming, monitoring, and incident response
Secure-by-design methodologies, frameworks, MLSecOps, and shift-left approaches
Mutuazione: 20830030 Cybersecurity and Artificial Intelligence in Ingegneria informatica e dell'intelligenza artificiale LM-32 IANNUCCI STEFANO, PIZZONIA MAURIZIO
Programme
Fundamentals of cybersecurity and machine learningOverview of the various contexts in which artificial intelligence can be used within an organization, including benefits and threats
Defensive use of artificial intelligence
Data formats: host-based and network-based
Feature engineering
Anomaly detection
Malware analysis
Network traffic analysis: data preparation, supervised learning, unsupervised learning, ensemble learning
Adversarial machine learning
AI-related threats and mitigation strategies
Model poisoning and tampering, defenses
Supply chain attacks and defenses
Evasion attacks and defenses
Privacy attacks and defenses
Generative AI: GANs for malicious applications (deepfakes, voice cloning, malware evasion, spam, video surveillance), deepfake detection
LLMs: architectures leveraging LLMs, RAG, MCP, chain-of-thought, threat models
Direct/indirect prompt injection, jailbreaks, phishing, style injection, role-playing, impersonation, ChatGPT_DAN, data exfiltration, privilege escalation, and remote code execution (e.g., through MCP), defenses and mitigations
Poisoning in RAG systems and fine-tuning. Supply chain risks: publishing a poisoned LLM. Benchmarks, red teaming, monitoring, and incident response
Secure-by-design methodologies, frameworks, MLSecOps, and shift-left approaches
Attendance
-Type of evaluation
Written exam, oral exam, project teacher profile teaching materials
Overview of the various contexts in which AI can be used within an organization, including benefits and threats.
# Defensive use of AI
Data formats: host-based and network-based
Feature engineering
Anomaly detection
Malware analysis
Network traffic analysis: data preparation, supervised learning, unsupervised learning, ensembles
Adversarial ML
# AI-related threats and mitigation strategies
Model poisoning and tampering, defenses
Supply chain attacks and defenses
Evasion attacks and defenses
Privacy attacks and defenses
Generative AI: GANs for malicious applications such as deepfakes, voice cloning, evasion, malware, spam, video surveillance; deepfake detection
LLMs: architectures leveraging LLMs, RAG, MCP, chain-of-thoughts, threat models
Direct and indirect prompt injection, jailbreaks, phishing, style injection, role-playing, impersonation, ChatGPT DAN, data exfiltration, privilege escalation and remote code execution, for example through MCP; defenses and mitigations
Poisoning in RAG and fine-tuning. Supply chain risks: publishing a poisoned LLM. Benchmarks. Red teaming. Monitoring, incident response.
Secure-by-design methodology, frameworks, MLSecOps, shift-left.
https://www.proquest.com/docview/2134682020/73CEC526C564F85PQ
- Adversarial AI Attacks, Mitigations, and Defense Strategies: A cybersecurity professional's guide to AI attacks, threat modeling, and securing AI with MLSecOps. John Sotiropoulos Sotiropoulos
https://www.proquest.com/docview/3081482260/3B9D9980BFF8481BPQ
Mutuazione: 20830030 Cybersecurity and Artificial Intelligence in Ingegneria informatica e dell'intelligenza artificiale LM-32 IANNUCCI STEFANO, PIZZONIA MAURIZIO
Programme
Cybersecurity and machine learning reviewOverview of the various contexts in which AI can be used within an organization, including benefits and threats.
# Defensive use of AI
Data formats: host-based and network-based
Feature engineering
Anomaly detection
Malware analysis
Network traffic analysis: data preparation, supervised learning, unsupervised learning, ensembles
Adversarial ML
# AI-related threats and mitigation strategies
Model poisoning and tampering, defenses
Supply chain attacks and defenses
Evasion attacks and defenses
Privacy attacks and defenses
Generative AI: GANs for malicious applications such as deepfakes, voice cloning, evasion, malware, spam, video surveillance; deepfake detection
LLMs: architectures leveraging LLMs, RAG, MCP, chain-of-thoughts, threat models
Direct and indirect prompt injection, jailbreaks, phishing, style injection, role-playing, impersonation, ChatGPT DAN, data exfiltration, privilege escalation and remote code execution, for example through MCP; defenses and mitigations
Poisoning in RAG and fine-tuning. Supply chain risks: publishing a poisoned LLM. Benchmarks. Red teaming. Monitoring, incident response.
Secure-by-design methodology, frameworks, MLSecOps, shift-left.
Core Documentation
- Machine Learning and Security : Protecting Systems with Data and Algorithms. Chio, Clarence; Freeman, Davidhttps://www.proquest.com/docview/2134682020/73CEC526C564F85PQ
- Adversarial AI Attacks, Mitigations, and Defense Strategies: A cybersecurity professional's guide to AI attacks, threat modeling, and securing AI with MLSecOps. John Sotiropoulos Sotiropoulos
https://www.proquest.com/docview/3081482260/3B9D9980BFF8481BPQ
Attendance
Students are encouraged to attend to classes.Type of evaluation
Project and discussion on the first part. Oral exam, or written exam depending on enrollment numbers, on the second part. teacher profile teaching materials
Overview of the various contexts in which artificial intelligence can be used within an organization, including benefits and threats
Defensive use of artificial intelligence
Data formats: host-based and network-based
Feature engineering
Anomaly detection
Malware analysis
Network traffic analysis: data preparation, supervised learning, unsupervised learning, ensemble learning
Adversarial machine learning
AI-related threats and mitigation strategies
Model poisoning and tampering, defenses
Supply chain attacks and defenses
Evasion attacks and defenses
Privacy attacks and defenses
Generative AI: GANs for malicious applications (deepfakes, voice cloning, malware evasion, spam, video surveillance), deepfake detection
LLMs: architectures leveraging LLMs, RAG, MCP, chain-of-thought, threat models
Direct/indirect prompt injection, jailbreaks, phishing, style injection, role-playing, impersonation, ChatGPT_DAN, data exfiltration, privilege escalation, and remote code execution (e.g., through MCP), defenses and mitigations
Poisoning in RAG systems and fine-tuning. Supply chain risks: publishing a poisoned LLM. Benchmarks, red teaming, monitoring, and incident response
Secure-by-design methodologies, frameworks, MLSecOps, and shift-left approaches
Mutuazione: 20830030 Cybersecurity and Artificial Intelligence in Ingegneria informatica e dell'intelligenza artificiale LM-32 IANNUCCI STEFANO, PIZZONIA MAURIZIO
Programme
Fundamentals of cybersecurity and machine learningOverview of the various contexts in which artificial intelligence can be used within an organization, including benefits and threats
Defensive use of artificial intelligence
Data formats: host-based and network-based
Feature engineering
Anomaly detection
Malware analysis
Network traffic analysis: data preparation, supervised learning, unsupervised learning, ensemble learning
Adversarial machine learning
AI-related threats and mitigation strategies
Model poisoning and tampering, defenses
Supply chain attacks and defenses
Evasion attacks and defenses
Privacy attacks and defenses
Generative AI: GANs for malicious applications (deepfakes, voice cloning, malware evasion, spam, video surveillance), deepfake detection
LLMs: architectures leveraging LLMs, RAG, MCP, chain-of-thought, threat models
Direct/indirect prompt injection, jailbreaks, phishing, style injection, role-playing, impersonation, ChatGPT_DAN, data exfiltration, privilege escalation, and remote code execution (e.g., through MCP), defenses and mitigations
Poisoning in RAG systems and fine-tuning. Supply chain risks: publishing a poisoned LLM. Benchmarks, red teaming, monitoring, and incident response
Secure-by-design methodologies, frameworks, MLSecOps, and shift-left approaches
Attendance
-Type of evaluation
Written exam, oral exam, project teacher profile teaching materials
Overview of the various contexts in which AI can be used within an organization, including benefits and threats.
# Defensive use of AI
Data formats: host-based and network-based
Feature engineering
Anomaly detection
Malware analysis
Network traffic analysis: data preparation, supervised learning, unsupervised learning, ensembles
Adversarial ML
# AI-related threats and mitigation strategies
Model poisoning and tampering, defenses
Supply chain attacks and defenses
Evasion attacks and defenses
Privacy attacks and defenses
Generative AI: GANs for malicious applications such as deepfakes, voice cloning, evasion, malware, spam, video surveillance; deepfake detection
LLMs: architectures leveraging LLMs, RAG, MCP, chain-of-thoughts, threat models
Direct and indirect prompt injection, jailbreaks, phishing, style injection, role-playing, impersonation, ChatGPT DAN, data exfiltration, privilege escalation and remote code execution, for example through MCP; defenses and mitigations
Poisoning in RAG and fine-tuning. Supply chain risks: publishing a poisoned LLM. Benchmarks. Red teaming. Monitoring, incident response.
Secure-by-design methodology, frameworks, MLSecOps, shift-left.
https://www.proquest.com/docview/2134682020/73CEC526C564F85PQ
- Adversarial AI Attacks, Mitigations, and Defense Strategies: A cybersecurity professional's guide to AI attacks, threat modeling, and securing AI with MLSecOps. John Sotiropoulos Sotiropoulos
https://www.proquest.com/docview/3081482260/3B9D9980BFF8481BPQ
Mutuazione: 20830030 Cybersecurity and Artificial Intelligence in Ingegneria informatica e dell'intelligenza artificiale LM-32 IANNUCCI STEFANO, PIZZONIA MAURIZIO
Programme
Cybersecurity and machine learning reviewOverview of the various contexts in which AI can be used within an organization, including benefits and threats.
# Defensive use of AI
Data formats: host-based and network-based
Feature engineering
Anomaly detection
Malware analysis
Network traffic analysis: data preparation, supervised learning, unsupervised learning, ensembles
Adversarial ML
# AI-related threats and mitigation strategies
Model poisoning and tampering, defenses
Supply chain attacks and defenses
Evasion attacks and defenses
Privacy attacks and defenses
Generative AI: GANs for malicious applications such as deepfakes, voice cloning, evasion, malware, spam, video surveillance; deepfake detection
LLMs: architectures leveraging LLMs, RAG, MCP, chain-of-thoughts, threat models
Direct and indirect prompt injection, jailbreaks, phishing, style injection, role-playing, impersonation, ChatGPT DAN, data exfiltration, privilege escalation and remote code execution, for example through MCP; defenses and mitigations
Poisoning in RAG and fine-tuning. Supply chain risks: publishing a poisoned LLM. Benchmarks. Red teaming. Monitoring, incident response.
Secure-by-design methodology, frameworks, MLSecOps, shift-left.
Core Documentation
- Machine Learning and Security : Protecting Systems with Data and Algorithms. Chio, Clarence; Freeman, Davidhttps://www.proquest.com/docview/2134682020/73CEC526C564F85PQ
- Adversarial AI Attacks, Mitigations, and Defense Strategies: A cybersecurity professional's guide to AI attacks, threat modeling, and securing AI with MLSecOps. John Sotiropoulos Sotiropoulos
https://www.proquest.com/docview/3081482260/3B9D9980BFF8481BPQ
Attendance
Students are encouraged to attend to classes.Type of evaluation
Project and discussion on the first part. Oral exam, or written exam depending on enrollment numbers, on the second part. teacher profile teaching materials
Overview of the various contexts in which artificial intelligence can be used within an organization, including benefits and threats
Defensive use of artificial intelligence
Data formats: host-based and network-based
Feature engineering
Anomaly detection
Malware analysis
Network traffic analysis: data preparation, supervised learning, unsupervised learning, ensemble learning
Adversarial machine learning
AI-related threats and mitigation strategies
Model poisoning and tampering, defenses
Supply chain attacks and defenses
Evasion attacks and defenses
Privacy attacks and defenses
Generative AI: GANs for malicious applications (deepfakes, voice cloning, malware evasion, spam, video surveillance), deepfake detection
LLMs: architectures leveraging LLMs, RAG, MCP, chain-of-thought, threat models
Direct/indirect prompt injection, jailbreaks, phishing, style injection, role-playing, impersonation, ChatGPT_DAN, data exfiltration, privilege escalation, and remote code execution (e.g., through MCP), defenses and mitigations
Poisoning in RAG systems and fine-tuning. Supply chain risks: publishing a poisoned LLM. Benchmarks, red teaming, monitoring, and incident response
Secure-by-design methodologies, frameworks, MLSecOps, and shift-left approaches
Mutuazione: 20830030 Cybersecurity and Artificial Intelligence in Ingegneria informatica e dell'intelligenza artificiale LM-32 IANNUCCI STEFANO, PIZZONIA MAURIZIO
Programme
Fundamentals of cybersecurity and machine learningOverview of the various contexts in which artificial intelligence can be used within an organization, including benefits and threats
Defensive use of artificial intelligence
Data formats: host-based and network-based
Feature engineering
Anomaly detection
Malware analysis
Network traffic analysis: data preparation, supervised learning, unsupervised learning, ensemble learning
Adversarial machine learning
AI-related threats and mitigation strategies
Model poisoning and tampering, defenses
Supply chain attacks and defenses
Evasion attacks and defenses
Privacy attacks and defenses
Generative AI: GANs for malicious applications (deepfakes, voice cloning, malware evasion, spam, video surveillance), deepfake detection
LLMs: architectures leveraging LLMs, RAG, MCP, chain-of-thought, threat models
Direct/indirect prompt injection, jailbreaks, phishing, style injection, role-playing, impersonation, ChatGPT_DAN, data exfiltration, privilege escalation, and remote code execution (e.g., through MCP), defenses and mitigations
Poisoning in RAG systems and fine-tuning. Supply chain risks: publishing a poisoned LLM. Benchmarks, red teaming, monitoring, and incident response
Secure-by-design methodologies, frameworks, MLSecOps, and shift-left approaches
Attendance
-Type of evaluation
Written exam, oral exam, project teacher profile teaching materials
Overview of the various contexts in which AI can be used within an organization, including benefits and threats.
# Defensive use of AI
Data formats: host-based and network-based
Feature engineering
Anomaly detection
Malware analysis
Network traffic analysis: data preparation, supervised learning, unsupervised learning, ensembles
Adversarial ML
# AI-related threats and mitigation strategies
Model poisoning and tampering, defenses
Supply chain attacks and defenses
Evasion attacks and defenses
Privacy attacks and defenses
Generative AI: GANs for malicious applications such as deepfakes, voice cloning, evasion, malware, spam, video surveillance; deepfake detection
LLMs: architectures leveraging LLMs, RAG, MCP, chain-of-thoughts, threat models
Direct and indirect prompt injection, jailbreaks, phishing, style injection, role-playing, impersonation, ChatGPT DAN, data exfiltration, privilege escalation and remote code execution, for example through MCP; defenses and mitigations
Poisoning in RAG and fine-tuning. Supply chain risks: publishing a poisoned LLM. Benchmarks. Red teaming. Monitoring, incident response.
Secure-by-design methodology, frameworks, MLSecOps, shift-left.
https://www.proquest.com/docview/2134682020/73CEC526C564F85PQ
- Adversarial AI Attacks, Mitigations, and Defense Strategies: A cybersecurity professional's guide to AI attacks, threat modeling, and securing AI with MLSecOps. John Sotiropoulos Sotiropoulos
https://www.proquest.com/docview/3081482260/3B9D9980BFF8481BPQ
Mutuazione: 20830030 Cybersecurity and Artificial Intelligence in Ingegneria informatica e dell'intelligenza artificiale LM-32 IANNUCCI STEFANO, PIZZONIA MAURIZIO
Programme
Cybersecurity and machine learning reviewOverview of the various contexts in which AI can be used within an organization, including benefits and threats.
# Defensive use of AI
Data formats: host-based and network-based
Feature engineering
Anomaly detection
Malware analysis
Network traffic analysis: data preparation, supervised learning, unsupervised learning, ensembles
Adversarial ML
# AI-related threats and mitigation strategies
Model poisoning and tampering, defenses
Supply chain attacks and defenses
Evasion attacks and defenses
Privacy attacks and defenses
Generative AI: GANs for malicious applications such as deepfakes, voice cloning, evasion, malware, spam, video surveillance; deepfake detection
LLMs: architectures leveraging LLMs, RAG, MCP, chain-of-thoughts, threat models
Direct and indirect prompt injection, jailbreaks, phishing, style injection, role-playing, impersonation, ChatGPT DAN, data exfiltration, privilege escalation and remote code execution, for example through MCP; defenses and mitigations
Poisoning in RAG and fine-tuning. Supply chain risks: publishing a poisoned LLM. Benchmarks. Red teaming. Monitoring, incident response.
Secure-by-design methodology, frameworks, MLSecOps, shift-left.
Core Documentation
- Machine Learning and Security : Protecting Systems with Data and Algorithms. Chio, Clarence; Freeman, Davidhttps://www.proquest.com/docview/2134682020/73CEC526C564F85PQ
- Adversarial AI Attacks, Mitigations, and Defense Strategies: A cybersecurity professional's guide to AI attacks, threat modeling, and securing AI with MLSecOps. John Sotiropoulos Sotiropoulos
https://www.proquest.com/docview/3081482260/3B9D9980BFF8481BPQ
Attendance
Students are encouraged to attend to classes.Type of evaluation
Project and discussion on the first part. Oral exam, or written exam depending on enrollment numbers, on the second part.