Exploring the Robustness of AI-Driven Tools in Digital Forensics: A Preliminary Study

Authors: Silvia Lucia Sanna, Leonardo Regano, Davide Maiorca, Giorgio Giacinto

Published: 2024-12-02 10:48:53+00:00

AI Summary

This preliminary study explores the robustness of commercial AI-driven digital forensics tools, Magnet AI and Excire Photo AI, against various non-standard and adversarial content. It evaluates their performance in classifying images related to nudity and deepfakes, as well as text chats, to assess their susceptibility to anti-forensics attacks. The research concludes that these AI algorithms are not robust enough, exhibiting limitations in accurately detecting and categorizing manipulated or ambiguous data.

Abstract

Nowadays, many tools are used to facilitate forensic tasks about data extraction and data analysis. In particular, some tools leverage Artificial Intelligence (AI) to automatically label examined data into specific categories (\\ie, drugs, weapons, nudity). However, this raises a serious concern about the robustness of the employed AI algorithms against adversarial attacks. Indeed, some people may need to hide specific data to AI-based digital forensics tools, thus manipulating the content so that the AI system does not recognize the offensive/prohibited content and marks it at as suspicious to the analyst. This could be seen as an anti-forensics attack scenario. For this reason, we analyzed two of the most important forensics tools employing AI for data classification: Magnet AI, used by Magnet Axiom, and Excire Photo AI, used by X-Ways Forensics. We made preliminary tests using about $200$ images, other $100$ sent in $3$ chats about pornography and teenage nudity, drugs and weapons to understand how the tools label them. Moreover, we loaded some deepfake images (images generated by AI forging real ones) of some actors to understand if they would be classified in the same category as the original images. From our preliminary study, we saw that the AI algorithm is not robust enough, as we expected since these topics are still open research problems. For example, some sexual images were not categorized as nudity, and some deepfakes were categorized as the same real person, while the human eye can see the clear nudity image or catch the difference between the deepfakes. Building on these results and other state-of-the-art works, we provide some suggestions for improving how digital forensics analysis tool leverage AI and their robustness against adversarial attacks or different scenarios than the trained one.

Key findings

The study found that the evaluated AI algorithms in commercial digital forensics tools are not robust enough, demonstrating high false positive rates and misclassifications for various content. For instance, some sexual images were not categorized as nudity, and deepfake images were often misidentified as real persons. This lack of robustness suggests that these tools could be evaded by sophisticated anti-forensics techniques.

Approach

The authors conducted preliminary tests on two commercial AI-driven digital forensics tools, Magnet AI (used by Magnet Axiom) and Excire Photo AI (used by X-Ways Forensics). They created custom datasets comprising 200 images for nudity detection (including real porn, teenager-appearing actors, painted nude dresses, and manga), 200 images for face recognition (popular actors, their deepfakes, and post-surgery faces), and three Facebook Messenger chats covering drugs, weapons, and sex-related content to evaluate the tools' classification and detection capabilities.

Datasets

Custom datasets created by the authors: 200 images for nude detection (real pornography, images of teenage-appearing actors, dresses with nude drawings, pornographic comic videos), 200 images for face recognition (popular actors and their deepfakes, pictures of their children, popular people before and after cosmetic or gender-affirming surgery), and 3 Facebook Messenger chats (drugs, weapons, sex) with over 100 messages each. Public datasets for weapons were mentioned (Qi et al., 2020), but chats and other image datasets were self-created.

Model(s)

UNKNOWN (The paper analyzes commercial black-box AI tools, Magnet AI and Excire Photo AI, without disclosing their underlying architectures. The authors did not use specific audio deepfake detection models in their main contribution, which focused on images and text.)

Author countries

Italy

← Previous