Cybersecurity and AI

Modern enterprises rely on cybersecurity tools—including endpoint, network, email, firewall, identity, and cloud-security systems—to detect attacks. When these tools cannot reach a confident conclusion, they generate alerts that must be prioritized and investigated by a Security Operations Center (SOC). This course shows how AI-enabled cybersecurity products can convert imperfect evidence into useful, auditable decision support for analysts. Students evaluate models, combine statistical and explicit evidence, account for uncertainty and review capacity, and design human-centered workflows. Suspicious-email triage provides the hands-on case, while the underlying architecture transfers across security domains and supports careers in AI, cybersecurity, and security-product development.

Prerequisites and Required Background 

Students should have the following preparation:

• Basic Python programming and the ability to read and modify short, scaffolded code segments.
• Introductory familiarity with supervised machine learning, classification, and common evaluation metrics.
• Comfort reading simple tables, plots, structured records, and model-evaluation output.
• No prior cybersecurity course or hands-on security-operations experience is required.

Primary Audience 

The course is intended primarily for Master of Science in AI students. Appropriately prepared CS or AI undergraduates may also participate. It is especially relevant to students interested in AI engineering for security products, cybersecurity analytics, detection engineering, technical product development, or other roles that connect machine learning with enterprise defensive operations.

Modern enterprises defend themselves through many layers of technology, including endpoint detection and response, network monitoring, email security, firewalls, identity and access management, cloud-security platforms, vulnerability-management systems, and malware-analysis tools. These systems observe different parts of enterprise activity and try to distinguish routine behavior from events that may require security attention. When the available evidence does not support a confident conclusion, they generate alerts that must be examined rather than treated automatically as attacks or harmless activity.

A Security Operations Center (SOC) receives, prioritizes, enriches, investigates, and escalates these alerts under limited time and analyst capacity. The central problem is therefore not simply whether an AI model achieves high accuracy. A useful cybersecurity product must help analysts decide what deserves attention, understand the evidence and uncertainty behind that recommendation, and identify an appropriate next step. Its outputs must remain understandable, auditable, and compatible with organizational policy and human oversight.

Students in this course take the role of developers and evaluators of AI-assisted cybersecurity products. They examine how model scores interact with incomplete or conflicting evidence, contextual information, asymmetric errors, thresholds, review capacity, explicit policy, structured explanations, and human review. The hands-on work follows one system from educational records through model evaluation, evidence and policy logic, analyst-facing outputs, and robustness testing. Students learn why a model score alone is not a complete security product.

Suspicious-email triage provides the concrete implementation thread, but email is a case study rather than the course’s limit. The same design principles apply to endpoint-alert triage, network-anomaly review, identity anomalies, vulnerability prioritization, malware analysis, and cloud-security findings. These transferable skills prepare students for work as AI or machine-learning engineers in security, cybersecurity-product developers, detection and analytics engineers, and technical evaluators or product leaders. Students will not become SOC analysts in five days, but they will learn enough about enterprise defensive workflows and analyst needs to design AI capabilities that fit real operational environments.

All exercises use inert educational records and supplied offline artifacts. Students will not access real inboxes, click live links, open attachments, query live threat-intelligence services, or take action in operational security systems.

By the end of the course, students will be able to:

CLO1 – Knowledge / Application: Explain how enterprise cybersecurity tools generate alerts and how SOCs triage uncertainty; identify and characterize security-relevant evidence, including its source, limits, uncertainty, and absence.

CLO2 – Analysis: Differentiate records, evidence, labels, model outputs, policy decisions, dispositions, actions, and outcomes.

CLO3 – Evaluation: Evaluate a defensive workflow using metrics, consequences, workload, scenario behavior, and robustness tests.

CLO4 – Creation / Synthesis: Design or adapt a bounded workflow combining a statistical model, explicit evidence, policy, structured output, and human review.

CLO5 – Evaluation / Judgment: Reason about oversight, escalation, auditability, monitoring, residual risk, feedback quality, and defensible claims.

This mapping uses the Master of Science in AI program learning outcomes supplied with the illustrative syllabus. It is proposed for academic review and should not be treated as approved until confirmed by Plaksha. 

H = high contribution; M = moderate contribution.

Proposed contribution summary: Primary PLOs – PLO2, PLO5, PLO6. Secondary PLOs – PLO1, PLO3.

Program Learning Outcomes Referenced Above 

PLO1: Demonstrate a solid understanding of AI fundamentals, mathematical foundations, algorithms, and data systems.

PLO2: Design, develop, and evaluate industry-grade AI and machine learning models for real-world applications.

PLO3: Apply MLOps principles and engineering practices to manage scalable AI systems.

PLO4: Deploy AI and ML systems effectively in cloud and production environments.

PLO5: Apply AI methods to support business strategy, decision-making, and domain-specific applications.

PLO6: Design and deliver comprehensive end-to-end AI solutions integrating data, models, infrastructure, and deployment.

The course comprises five three-hour teaching days. Sunday, December 20, 2026 is not a class day. Each meeting combines instructor content, a bounded demonstration, hands-on student work, and a debrief or checkoff. Individual in-class conceptual quizzes are administered on Day 3 and Day 5.

Daily Engagement Structure 

Each lecture day includes approximately three contact hours. The exact daily start and end times will be confirmed by Plaksha. The course protects substantial in-class work time because every graded lecture-day artifact depends on supervised analysis, demonstrations, and cumulative use of the educational course package. Under normal course arrangements, all graded work, including both conceptual quizzes, is completed in person during the scheduled class meetings; an alternative arrangement applies only when required by an officially approved Plaksha University policy or accommodation.

The course includes five in-class practical or analytical assessments and two in-class conceptual quizzes. Days 1, 2, and 5 and both quizzes are individual; Days 3 and 4 combine pair work with individual accountability. Together these seven components compose 100% of the final grade. There are no separate conventional homework assignments.

• Day 1 Individual Evidence-and-Label Analysis — 10% 

  Students analyze a set of educational records, distinguish represented from missing evidence, interpret the educational label within its limits, and state what the record does not justify claiming.

• Day 2 Individual Fixed-Model Evaluation — 15% 

  Students calculate and interpret model-evaluation results at the frozen operating points, verify denominators, compare standard and deployment-like populations, analyze assigned errors, and explain threshold, prevalence, workload, and consequence tradeoffs.

• Day 3 Pair Evidence-and-Policy Implementation — 15% 

  Pairs complete bounded identity-evidence, request-consequence, and policy logic within supplied scaffolding and run visible tests. Thirteen points are pair-shared and two points are earned through individual understanding checks.

• Day 3 In-Class Conceptual Quiz — 7.5% 

  This individual quiz is completed during the Day 3 class meeting and assesses conceptual material from the instructor lectures. It requires no programming. The exact format and coverage will be announced by the instructor.

• Day 4 Pair Integrated-Product Evaluation — 30% 

  Pairs evaluate the integrated product across policy profiles, structured records, errors, review priority, bounded next steps, and a supplied human-review event. Twenty-six points are pair-shared. Four points are individually attributable: a three-point structured checkoff plus one point for contribution-and-traceability evidence.

• Day 5 Individual Robustness Analysis — 15% 

  Students analyze static challenge outputs, identify the first layer at which behavior diverges, trace downstream effects, propose a bounded control and revalidation plan, state residual risk, and rewrite an overbroad claim. Challenge data may not be used to repair or tune the system.

• Day 5 In-Class Conceptual Quiz — 7.5% 

  This individual quiz is completed during the Day 5 class meeting and assesses conceptual material from the instructor lectures. It requires no programming. The exact format and coverage will be announced by the instructor.

Individual accountability: 61% of the final grade is directly attributable to individual performance; 39% is pair-shared. The five practical or analytical assessments account for 85% of the grade, and the two individual in-class conceptual quizzes account for 15%. All graded work is completed in class. There is no separate attendance or participation percentage.

Evaluation Principles 

• Technical calculations, evidence, assumptions, workload, consequences, and tradeoffs must be stated explicitly.
• Alternative recommendations may earn full credit when they are consistent with the represented evidence and are defended clearly.
• Educational outputs must not be presented as determinations of phishing, maliciousness, attacker intent, compromise, safety, or incident status.
• Students may use either a complete eligible student implementation or a complete reference continuation when authorized; hybrid implementations are not permitted.

Active participation is required because each meeting integrates lecture, analysis, demonstration, graded hands-on work, and, on Days 3 and 5, an in-class conceptual quiz. Participation includes preparation, careful reasoning, constructive pair collaboration, and engagement in debriefs. It is not assigned a separate percentage because the five practical or analytical assessments and two quizzes already compose 100% of the grade.

Students should expect approximately four to six hours of work outside scheduled class meetings over the entire course. This time is primarily for pre-course software and environment verification, reviewing lecture concepts in preparation for the two in-class quizzes, briefly reviewing instructions or organizing files, and limited pair coordination where necessary.

There are no separate conventional homework assignments. No substantial software development, major overnight coding, open-ended model development or tuning, lengthy report writing, or prolonged at-home debugging is intended. Daily practical and analytical assessments are designed to be completed during the scheduled class meeting.

Attendance at all five class meetings is required. This is an intensive one-credit course with only five lecture days. Each meeting combines instructor lectures, demonstrations, supervised practical work, individual or pair assessment, and, on Days 3 and 5, an in-class conceptual quiz. The five practical or analytical assessments and the two quizzes are all conducted during scheduled class meetings and together constitute 100% of the course grade. The grading weight assigned to each lecture day and quiz is listed earlier in the Class Schedule, Assignments, and Grading sections. Students cannot earn the associated assessment credit without attending the class meeting and completing the required in-class work, except where an officially approved Plaksha University policy or accommodation requires an alternative arrangement. A student facing an unavoidable circumstance should contact the instructor promptly and follow the applicable Plaksha process.

The preferred contact method is email at barua@umd.edu. Meetings are available by appointment. Consultation arrangements during the course period, including any in-person availability before or after class, will be announced at the beginning of the course.

The instructional content, lecture dates, contact hours, and assessment design are complete. The items below are administrative or institutional-policy fields that Plaksha may supply or confirm during approval.

_{Prepared for Plaksha University Board of Studies review | Revised August 2026}