The rapid development of artificial intelligence (AI), particularly Large Language Models (LLMs), has triggered a new generation of social Sybil attacks. Considering that 74% of Ukrainians use social media as their primary source of information, this poses unprecedented threats to cybersecurity and the integrity of online communication. Modern AI bot networks, capable of perfectly mimicking human behavior, are actively used to spread disinformation and manipulate public opinion. This paper analyzes existing methods for Sybil attack detection—including graph-based, behavioral, and linguistic approaches—and demonstrates their growing ineffectiveness against bots enhanced by the generative capabilities of LLMs. The analysis of recent research shows that traditional detectors, which relied on profile metadata, linguistic verification, and social graph anomalies, are no longer reliable. Modern botnets, such as the "fox8" network discovered in 2023, have learned to mask metadata, generate stylistically rich content, and imitate organic social connections. This threat is compounded by the fact that social media users correctly identify bots in only 42% of cases, while AI-generated propaganda receives 37% more engagement than content created by humans. This article systematizes new countermeasures, including the use of LLMs themselves to detect stylistic anomalies in text (e.g., perplexity analysis) and tests based on cognitive asymmetries. Promising future research directions include the development of multimodal detectors, the creation of autonomous, self-updating systems, and a shift in focus from detecting individual bots to identifying coordinated manipulative campaigns. Consequently, a fundamental reassessment of detection approaches is one of today's most critical challenges.
O.P. Melnychuk (Wed,) studied this question.