Predictive cybersecurity based on artificial intelligence against generative attacks

Abstract

The study analyzes predictive cybersecurity based on artificial intelligence as a response to the growing emergence of generative attacks capable of producing hyper-personalized phishing, deepfakes, synthetic identities, adaptive malware, and increasingly sophisticated evasion strategies. The objective was to examine recent literature regarding advances, risks, limitations, and defensive implementation criteria against these threats. Methodologically, a bibliographic review was conducted using scientific articles, technical reports, and specialized frameworks, without collecting personal data or performing offensive testing. The findings reveal that artificial intelligence strengthens early threat detection by correlating weak signals, identifying multi-source anomalies, and partially overcoming the limitations of static signature-based defenses. However, it also presents challenges related to data bias, false positives, algorithmic opacity, adversarial manipulation, prompt injection, and excessive reliance on automation. The study concludes that predictive defense should be integrated into layered security architectures supported by data governance, operational explainability, continuous adversarial evaluation, human supervision, and institutional response protocols in order to transition from reactive cybersecurity toward an anticipatory, resilient, and responsible model.