Javascript must be enabled to continue!

Adversarial Machine Learning: Attack Vectors, Defences, and Robustness

Background. Adversarial machine learning has progressed from a marginal concern within machine learning research into a first-order discipline for the secure deployment of artificial intelligence systems in regulated and operational environments. The contemporary threat landscape encompasses evasion at inference time, data poisoning across training pipelines, model extraction and inference attacks against deployed systems, and a defence ecosystem whose claimed robustness frequently fails to generalise beyond the threat models under which it was measured. Purpose. This paper synthesises the adversarial machine learning literature spanning 2012 to 2025 to provide practitioners with an analytically grounded mapping between attack taxonomies, defence mechanisms, and the governance and risk-management controls required for defensible AI deployment. It further examines a structural critique of current robustness benchmarking practice and its implications for procurement, assurance, and regulatory compliance. Approach. The paper adopts a narrative literature review methodology, drawing on authoritative primary sources, including NIST AI 100-2 E2025, the OWASP Top 10 for Large Language Model Applications, MITRE ATLAS, and peer-reviewed research on adversarial machine learning. Sources were selected for authority, currency, and direct relevance to operational practice across both predictive and generative AI systems. Findings. Three findings are advanced. First, transferability renders the black-box assumption operationally weak, requiring defenders to assume adversarial knowledge of model behaviour as the default threat posture. Second, supply chain risks for datasets, pre-trained models, and fine-tuning pipelines are now first-order concerns that map to existing governance, risk, and compliance primitives but require AI-specific controls. Third, robustness benchmarking practices systematically overestimate defensive efficacy, with implications for procurement decisions and regulatory assurance. Implications. Practitioners require a layered control architecture combining adversarial training, certified components on highest-risk decision paths, input validation, continuous red-teaming, and governance treatments that operationalise threat models within ISO/IEC 42001 and the NIST AI Risk Management Framework. Procurement, vendor onboarding, and assurance processes should integrate adversarial robustness evaluation against named threat models rather than relying on aggregate robustness claims. Keywords: adversarial machine learning; AI security; evasion attacks; data poisoning; model extraction; certified robustness; AI governance JEL Classification: O33; K24; L86 ACM CCS: Security and privacy ~ Software and application security; Computing methodologies ~ Machine learning ~ Adversarial learning; Security and privacy ~ Human and societal aspects of security and privacy ~ Governance

Elsevier BV

Rizwan Tanveer

2026

Title: Adversarial Machine Learning: Attack Vectors, Defences, and Robustness

Description:

Background.

Adversarial machine learning has progressed from a marginal concern within machine learning research into a first-order discipline for the secure deployment of artificial intelligence systems in regulated and operational environments.

The contemporary threat landscape encompasses evasion at inference time, data poisoning across training pipelines, model extraction and inference attacks against deployed systems, and a defence ecosystem whose claimed robustness frequently fails to generalise beyond the threat models under which it was measured.

Purpose.

This paper synthesises the adversarial machine learning literature spanning 2012 to 2025 to provide practitioners with an analytically grounded mapping between attack taxonomies, defence mechanisms, and the governance and risk-management controls required for defensible AI deployment.

It further examines a structural critique of current robustness benchmarking practice and its implications for procurement, assurance, and regulatory compliance.

Approach.

The paper adopts a narrative literature review methodology, drawing on authoritative primary sources, including NIST AI 100-2 E2025, the OWASP Top 10 for Large Language Model Applications, MITRE ATLAS, and peer-reviewed research on adversarial machine learning.

Sources were selected for authority, currency, and direct relevance to operational practice across both predictive and generative AI systems.

Findings.

Three findings are advanced.

First, transferability renders the black-box assumption operationally weak, requiring defenders to assume adversarial knowledge of model behaviour as the default threat posture.

Second, supply chain risks for datasets, pre-trained models, and fine-tuning pipelines are now first-order concerns that map to existing governance, risk, and compliance primitives but require AI-specific controls.

Third, robustness benchmarking practices systematically overestimate defensive efficacy, with implications for procurement decisions and regulatory assurance.

Implications.

Practitioners require a layered control architecture combining adversarial training, certified components on highest-risk decision paths, input validation, continuous red-teaming, and governance treatments that operationalise threat models within ISO/IEC 42001 and the NIST AI Risk Management Framework.

Procurement, vendor onboarding, and assurance processes should integrate adversarial robustness evaluation against named threat models rather than relying on aggregate robustness claims.

Keywords: adversarial machine learning; AI security; evasion attacks; data poisoning; model extraction; certified robustness; AI governance JEL Classification: O33; K24; L86 ACM CCS: Security and privacy ~ Software and application security; Computing methodologies ~ Machine learning ~ Adversarial learning; Security and privacy ~ Human and societal aspects of security and privacy ~ Governance.

Back

Malware threatens cybersecurity by enabling data theft, unauthorized access, and extortion. Traditional malware detection systems (MDS) struggle with the increasing volume and comp...

Adversarial Robustness Improvement for Deep Neural Networks

Abstract Deep neural networks (DNNs) are key components for the implementation of autonomy in systems that operate in highly complex and unpredictable environments (self-dr...

Adversarial examples attack based on random warm restart mechanism and improved Nesterov momentum

The deep learning algorithm has achieved great success in the field of computer vision, but some studies have pointed out that the deep learning model is vulnerable to attacks adve...

Adversarial examples attack based on random warm restart mechanism and improved Nesterov momentum

The deep learning algorithm has achieved great success in the field of computer vision, but some studies have pointed out that the deep learning model is vulnerable to attacks adve...

Efficient Defense Against First Order Adversarial Attacks on Convolutional Neural Networks

Machine learning models, especially neural networks, are vulnerable to adversarial attacks, where inputs are purposefully altered to induce incorrect predictions. These adversarial...

Adversarial Training and Robustness in Machine Learning Frameworks

In the realm of machine learning, ensuring robustness against adversarial attacks is increasingly crucial. Adversarial training has emerged as a prominent strategy to fortify model...

Mitigating Adversarial Attacks Uncertainty Through Interval Analysis

Abstract The adversarial attack is characterized by a high attack success rate and a fast generation of examples. It is widely used in neural network robustness eva...

Selection of Injectable Drug Product Composition using Machine Learning Models (Preprint)

BACKGROUND As of July 2020, a Web of Science search of “machine learning (ML)” nested within the search of “pharmacokinetics or pharmacodynamics” yielded over 100...

Email:
Password:

Email:

Adversarial Machine Learning: Attack Vectors, Defences, and Robustness

Related Results