EUVIMEDEUVIMED entdecken
Crossref

AI Autonomy – Can Humanity Feel Threatened?Analyzing the Technical Limitations and Agency Potential of Large Language Models

Krzysztof Jan Jakubski

Cybersecurity & Cybercrime · 04.05.2026 · 1 (10) · 339-355

DOI: 10.5604/01.3001.0055.7867 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Cybersecurity & Cybercrime, erschienen am 04.05.2026. Verfasst von Krzysztof Jan Jakubski. Die Arbeit nennt 36 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This article analyzes the discrepancy between sensational media reports about the "rebellion" of artificial intelligence and the actual technical capabilities of large language models (LLMs). Drawing on the relevant literature, it is demonstrated that attributing intentionality to models is an anthropomorphic fallacy, and that the observed anomalies (such as alignment faking) result from statistical optimization. Further, drawing on security and criminological analyses, the article identifies the "physical barrier" as a key limitation preventing AI from posing an existential threat to humanity at its current stage of development, and defines the real threat of technocrime (AIC), where AI acts as an advanced tool rather than an autonomous perpetrator.

Zitieren

APA

Jakubski, K. J. (2026). AI Autonomy – Can Humanity Feel Threatened?Analyzing the Technical Limitations and Agency Potential of Large Language Models. Cybersecurity & Cybercrime, 1(10), 339-355. https://doi.org/10.5604/01.3001.0055.7867

Vancouver

Jakubski KJ. AI Autonomy – Can Humanity Feel Threatened?Analyzing the Technical Limitations and Agency Potential of Large Language Models. Cybersecurity & Cybercrime. 2026;1(10):339-355. doi:10.5604/01.3001.0055.7867

BibTeX
@article{Jakubski2026,
  author = {Jakubski, Krzysztof Jan},
  title = {AI Autonomy – Can Humanity Feel Threatened?Analyzing the Technical Limitations and Agency Potential of Large Language Models},
  journal = {Cybersecurity & Cybercrime},
  year = {2026},
  volume = {1},
  number = {10},
  pages = {339--355},
  doi = {10.5604/01.3001.0055.7867}
}
Kostenlos starten →