Mirko Casu

Roberto Leotta, Salvatore Alfio Sambataro, Claudio Vittorio Ragaglia et al.

The landscape of synthetic media has been irrevocably altered by text-to-video (T2V) models, whose outputs are rapidly approaching indistinguishability from reality. Critically, this technology is no longer confined to large-scale labs; the proliferation of efficient, open-source generators is democratizing the ability to create high-fidelity synthetic content on consumer-grade hardware. This makes existing face-centric and manipulation-based benchmarks obsolete. To address this urgent threat, we introduce SynthForensics, to the best of our knowledge the first human-centric benchmark for detecting purely synthetic video deepfakes. The benchmark comprises 6,815 unique videos from five architecturally distinct, state-of-the-art open-source T2V models. Its construction was underpinned by a meticulous two-stage, human-in-the-loop validation to ensure high semantic and visual quality. Each video is provided in four versions (raw, lossless, light, and heavy compression) to enable real-world robustness testing. Experiments demonstrate that state-of-the-art detectors are both fragile and exhibit limited generalization when evaluated on this new domain: we observe a mean performance drop of $29.19\%$ AUC, with some methods performing worse than random chance, and top models losing over 30 points under heavy compression. The paper further investigates the efficacy of training on SynthForensics as a means to mitigate these observed performance gaps, achieving robust generalization to unseen generators ($93.81\%$ AUC), though at the cost of reduced backward compatibility with traditional manipulation-based deepfakes. The complete dataset and all generation metadata, including the specific prompts and inference parameters for every video, will be made publicly available at [link anonymized for review].

Pietro Bongini, Sara Mandelli, Andrea Montibeller et al.

Synthetic image source attribution is an open challenge, with an increasing number of image generators being released yearly. The complexity and the sheer number of available generative techniques, as well as the scarcity of high-quality open source datasets of diverse nature for this task, make training and benchmarking synthetic image source attribution models very challenging. WILD is a new in-the-Wild Image Linkage Dataset designed to provide a powerful training and benchmarking tool for synthetic image attribution models. The dataset is built out of a closed set of 10 popular commercial generators, which constitutes the training base of attribution models, and an open set of 10 additional generators, simulating a real-world in-the-wild scenario. Each generator is represented by 1,000 images, for a total of 10,000 images in the closed set and 10,000 images in the open set. Half of the images are post-processed with a wide range of operators. WILD allows benchmarking attribution models in a wide range of tasks, including closed and open set identification and verification, and robust attribution with respect to post-processing and adversarial attacks. Models trained on WILD are expected to benefit from the challenging scenario represented by the dataset itself. Moreover, an assessment of seven baseline methodologies on closed and open set attribution is presented, including robustness tests with respect to post-processing.

Mirko Casu, Luca Guarnera, Pasquale Caponnetto et al.

This paper examines the impact of cognitive biases on decision-making in forensics and digital forensics, exploring biases such as confirmation bias, anchoring bias, and hindsight bias. It assesses existing methods to mitigate biases and improve decision-making, introducing the novel "Impostor Bias", which arises as a systematic tendency to question the authenticity of multimedia content, such as audio, images, and videos, often assuming they are generated by AI tools. This bias goes beyond evaluators' knowledge levels, as it can lead to erroneous judgments and false accusations, undermining the reliability and credibility of forensic evidence. Impostor Bias stems from an a priori assumption rather than an objective content assessment, and its impact is expected to grow with the increasing realism of AI-generated multimedia products. The paper discusses the potential causes and consequences of Impostor Bias, suggesting strategies for prevention and counteraction. By addressing these topics, this paper aims to provide valuable insights, enhance the objectivity and validity of forensic investigations, and offer recommendations for future research and practical applications to ensure the integrity and reliability of forensic practices.

Claudio Giusti, Luca Guarnera, Mirko Casu et al.

Detecting fraudulent credit card transactions remains a significant challenge, due to the extreme class imbalance in real-world data and the often subtle patterns that separate fraud from legitimate activity. Existing research commonly attempts to address this by generating synthetic samples for the minority class using approaches such as GANs, VAEs, or hybrid generative models. However, these techniques, particularly when applied only to minority-class data, tend to result in overconfident classifiers and poor latent cluster separation, ultimately limiting real-world detection performance. In this study, we propose the Causal Prototype Attention Classifier (CPAC), an interpretable architecture that promotes class-aware clustering and improved latent space structure through prototype-based attention mechanisms and we will couple it with the encoder in a VAE-GAN allowing it to offer a better cluster separation moving beyond post-hoc sample augmentation. We compared CPAC-augmented models to traditional oversamplers, such as SMOTE, as well as to state-of-the-art generative models, both with and without CPAC-based latent classifiers. Our results show that classifier-guided latent shaping with CPAC delivers superior performance, achieving an F1-score of 93.14\% percent and recall of 90.18\%, along with improved latent cluster separation. Further ablation studies and visualizations provide deeper insight into the benefits and limitations of classifier-driven representation learning for fraud detection. The codebase for this work will be available at final submission.