Generative Models for Security: Attacks, Defenses, and Opportunities

View on arXiv ← Back to list

Authors: Luke A. Bauer, Vincent Bindschaedler

Published: 2021-07-21 15:16:10+00:00

AI Summary

This paper provides a comprehensive survey of recent research at the intersection of generative models and security/privacy. It explores how generative models are utilized in adversarial machine learning, for enhancing attacks (e.g., biometrics spoofing, malware obfuscation), and as building blocks for defenses (e.g., intrusion detection, privacy-preserving data synthesis, steganography, fairness). The survey also highlights the emerging threats from generative models, such as the creation of synthetic media like deepfakes for disinformation.

Abstract

Generative models learn the distribution of data from a sample dataset and can then generate new data instances. Recent advances in deep learning has brought forth improvements in generative model architectures, and some state-of-the-art models can (in some cases) produce outputs realistic enough to fool humans. We survey recent research at the intersection of security and privacy and generative models. In particular, we discuss the use of generative models in adversarial machine learning, in helping automate or enhance existing attacks, and as building blocks for defenses in contexts such as intrusion detection, biometrics spoofing, and malware obfuscation. We also describe the use of generative models in diverse applications such as fairness in machine learning, privacy-preserving data synthesis, and steganography. Finally, we discuss new threats due to generative models: the creation of synthetic media such as deepfakes that can be used for disinformation.

Key findings
The survey reveals that generative models are a dual-use technology, both enhancing various cybersecurity attacks and serving as foundational elements for robust defenses. A significant finding is the escalating threat of highly realistic synthetic media, including deepfakes, which complicates information trust and necessitates continuous research into detection and attribution. The paper identifies evaluating generative models and effectively detecting and attributing synthetic media as major open research challenges.
Approach
The authors conducted a comprehensive literature review, structuring their survey around the applications of generative models in security and privacy. They categorize these uses into attacks and defenses, covering areas such as intrusion detection, biometrics, malware, data synthesis, adversarial machine learning, steganography, and fairness. Additionally, they discuss the threats posed by generated synthetic media and outline key open research problems.
Datasets
UNKNOWN
Model(s)
UNKNOWN
Author countries
USA
← Previous
Next →