CyberLens

Week 2 of CYBR 325 introduced the foundations of artificial intelligence. Dr. Cooper explained how artificial intelligence, machine learning, and deep learning relate to one another, traced the field’s history, and introduced algorithms including decision trees and k-nearest neighbors. Our readings included Haenlein and Kaplan’s history of AI and IBM’s overview of artificial intelligence. In the breakout rooms, my group was assigned gradient descent and genetic algorithms. I wanted a source that connected these mechanics to a security question: what happens when someone attacks the learning process?

I chose Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations, published by NIST on March 24, 2025. The report organizes attacks by the system targeted, the machine learning life cycle stage, and the attacker’s objectives, capabilities, and knowledge. It distinguishes predictive AI from generative AI and examines threats including evasion, poisoning, privacy attacks, and prompt injection (Vassilev et al., 2025).

The life cycle perspective is connected directly to my breakout topic. Gradient descent adjusts a model’s parameters during training to reduce a loss function. It is an optimization method, not inherently a vulnerability. However, if an adversary manipulates training data, the learning process may incorporate patterns that support the attacker’s goals. The optimizer can function as designed while the resulting model behaves incorrectly. Successful poisoning is not guaranteed, but the possibility shows why protecting training data matters.

Genetic algorithms suggested another connection. They explore possible solutions through variation and selection guided by a fitness measure. I could imagine an attacker using a similar process to search for modified malware samples that evade a classifier while retaining their malicious functionality. This is my own connection to the course material, rather than an example presented this way by NIST. It helped me understand evasion as a search problem.

NIST’s attacker objectives also connect to familiar security concerns, including availability, integrity, and privacy. For generative AI, the report adds misuse enablement. This reminded me of the CIA triad from CYBR 332, although privacy is broader than confidentiality. The familiar principles still matter, but their application expands to training data, model parameters, and AI-generated outputs.

IBM’s overview reinforces this connection by identifying data poisoning and unauthorized changes to model architecture, weights, or parameters as AI risks (Stryker & Kavlakoglu, 2026). For a cybersecurity model, protecting integrity therefore includes examining how its data and components were obtained.

My main takeaway is that understanding AI fundamentals helps explain its security risks. Learning how optimization works made poisoning easier to understand, while genetic algorithms gave me a way to think about an attacker’s search for evasive inputs. Pairing a threat taxonomy with a fundamentals week helped me connect the components discussed in class to the ways an adversary might target them. I expect to revisit this report when we study risk management and governance.

References

Haenlein, M., & Kaplan, A. (2019). A brief history of artificial intelligence: On the past, present, and future of artificial intelligence. California Management Review, 61(4), 5–14. https://doi.org/10.1177/0008125619864925

Stryker, C., & Kavlakoglu, E. (2026, June 15). What is artificial intelligence (AI)? IBM. https://www.ibm.com/think/topics/artificial-intelligence

Vassilev, A., Oprea, A., Fordyce, A., Anderson, H., Davies, X., & Hamin, M. (2025). Adversarial machine learning: A taxonomy and terminology of attacks and mitigations (NIST AI 100-2e2025). National Institute of Standards and Technology. https://doi.org/10.6028/NIST.AI.100-2e2025

AI use disclosure

Claude (Anthropic) was used at Level 3, AI Collaboration, to assist with source discovery, organization, drafting, and revision. ChatGPT (OpenAI) assisted with checking source details, identifying technical overstatements, and revising wording and references. I reviewed the sources and contributed the course connections and personal analysis.