Digital communication channels have become major vectors for large-scale cyber fraud, including spam, phishing, and recruitment scams, causing significant financial losses and eroding user trust. While transformer-based models have improved text classification, most systems remain English-centric, limiting their effectiveness in multilingual environments and across diverse fraud typologies. In this work, we hypothesize that fraud detection models trained in a single language can exhibit cross-lingual generalization capabilities. We investigate the use of Natural Language Processing (NLP) for detecting multiple forms of textual fraud by developing and evaluating three transformer-based discriminative models targeting spam, phishing, and fake job postings. Specifically, we compare two architectures—MrBERT (308M parameters) and mROBERTa (283M parameters)—across English, Spanish, and Valencian data. Our results demonstrate that while models trained exclusively on English achieve near-perfect performance in-language (F1 ≈ 0.98), they exhibit limited cross-lingual generalization in other languages such as Spanish (F1 ≈ 0.44 for phishing detection, and F1 ≈ 0.85 for spam detection). However, incorporating multilingual training data dramatically improves performance in target languages (F1 ≈ 0.95–0.99) while maintaining or even enhancing English accuracy. We further observe that cross-lingual transfer is most effective when datasets are parallel or closely aligned. These findings underscore the critical importance of multilingual data inclusion for building robust, scalable cybersecurity frameworks in diverse linguistic settings.

The detection of toxic content in social media has become a critical task in Natural Language Processing (NLP), particularly given its intersection with complex issues like subjectivity, implicit language, and cultural context. Among these challenges, bias in training data remains a central concern—especially as language models risk reproducing and amplifying societal inequalities. This paper investigates the interplay between toxicity and gender bias on Twitter/X by introducing a novel dataset of violent and non-violent tweets, annotated not only for violence but also for gender. We conduct an exploratory analysis of how biased data can distort toxicity classification and present algorithms to mitigate these effects through dataset balancing and debiasing. Our contributions include four new dataset splits—two balanced and two debiased—that aim to support the development of fairer and more inclusive NLP models. By foregrounding the importance of equity in data curation, this work lays the groundwork for more ethical approaches to automated violence detection and gender annotation.

Image captioning bridges computer vision and natural language processing but remains vulnerable to social biases. This study evaluates gender bias in ChatGPT, Copilot, and Grok by analyzing their descriptions of fashion-related images prompted without gender cues. We introduce a methodology combining gender annotation, stereotype classification, and a manually curated dataset. Results show that GPT-4o and Grok frequently assign gender and reinforce stereotypes, while Copilot more often generates neutral captions. Grok shows the lowest error rate but consistently assigns gender, even when cues are ambiguous. These findings highlight the need for bias-aware captioning approaches in multimodal systems.