arXiv:2606.06879v1 Announce Type: new Abstract: Our prior work introduced COVA, a synthetically generated multi-turn conversational smishing dataset of 3,201 labeled conversations, establishing baseline detection benchmarks across eight models. While XGBoost with TF-IDF features achieved the best performance, with 72.5\% accuracy and 0.691 macro F1, transformer models underperformed, which was attributed to input truncation and insufficient training data. We present COVA-X, an expanded dataset of 10,985 conversations spanning eight elder-targeted scam categories, produced by an improved genera
The continuous evolution of AI models and the increasing sophistication of scams necessitate improved detection datasets, reflecting ongoing efforts to combat cybercrime.
This expanded dataset significantly improves the ability to detect multi-turn smishing attacks, particularly those targeting vulnerable populations like the elderly, by providing more robust training data for AI models.
The availability of COVA-X allows for the development and deployment of more accurate and resilient AI-powered smishing detection systems, reducing the impact of these social engineering attacks.
- · Cybersecurity firms
- · Elderly population
- · Financial institutions
- · AI researchers
- · Scammers
- · Cybercriminals
Improved detection rates for conversational smishing will lead to fewer successful scam attempts.
The reduced effectiveness of smishing may prompt scammers to develop new, more complex social engineering tactics or shift to different attack vectors.
This arms race could drive further innovation in AI-driven threat detection, potentially leading to more generalized AI agents capable of identifying novel scam methodologies.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.CL