20.500.12556/DKUM-97816 Spear phishing in veliki jezikovni modeli vpliv na varnostno ozaveščenost uporabnikov Spear phishing and large language models: Impact on user security awareness diplomsko delo visokošolskega študijskega programa Informacijska varnost V diplomski nalogi je bil izveden sistematični pregled literature o spear phishingu, velikih jezikovnih modelih (large language model, LLM) in varnostni ozaveščenosti uporabnikov. Namen je bil pokazati, kako LLM vplivajo na nastanek in prepričljivost ciljanih prevar ter katere obrambe in vrste izobraževanja najbolje delujejo. Gradivo je bilo iskano v bazah Scopus, Web of Science in IEEE Xplore; vključene so bile recenzirane objave iz let 2010–2025 (s poudarkom 2023–2025). Ugotovili smo, da LLM znižajo strošek in čas priprave napadov ter izboljšajo jezik in ton sporočil, zato so le-ta težje prepoznavna, tudi v SMS. Pokazano je bilo, da uporabniki pogosto precenijo svojo sposobnost prepoznavanja, kar poveča tveganje. Najboljše rezultate prinašajo kratki, redni in po vlogah prilagojeni programi ozaveščanja, povezani s konkretnimi koraki (npr. preverjanje po drugem kanalu) in krepitvijo samoučinkovitosti, samo opozarjanje na grožnjo ni zadostovalo. Kot učinkovita se je pokazala večslojna obramba: avtentikacija e-pošte (SPF, DKIM, DMARC), večfaktorska prijava, hibridna detekcija ter jasen in kratek načrt odziva. A systematic literature review on spear phishing, large language models (LLMs), and user security awareness was conducted. The aim was to show how LLMs affect the creation and persuasiveness of targeted scams and which defenses and training approaches work best. Material was searched in Scopus, Web of Science, and IEEE Xplore; peer-reviewed publications from 2010–2025 were included, with emphasis on 2023–2025. It was found that LLMs lower the cost and time of preparing attacks and improve language and tone, making messages harder to detect, including over SMS. Users were found to often overestimate their detection ability, which increases risk. The best results were achieved by short, regular, role-specific awareness programs paired with concrete steps (e.g., out-of-band verification) and strengthened self-efficacy; mere warnings about threats were not sufficient. Effective protection was shown to be multilayered: email authentication (SPF, DKIM, DMARC), multi-factor authentication, hybrid detection, and a clear, concise incident response plan. informacijska varnost varnostna ozaveščenost spear phishing veliki jezikovni modeli socialni inženiring diplomske naloge spear phishing phishing bulk phishing clone phishing vishing smishing social engineering targeted attacks cybersecurity information security security awareness user awareness human factor user vulnerability optimism bias self-efficacy perceived vulnerability psychological factors user education security training SETA fear appeal Protection Motivation Theory phishing simulations incident reporting email fraudulent emails credential theft malicious links scam detection phishing detection hybrid detection machine learning LLM large language models artificial intelligence generative artificial intelligence ChatGPT Claude Gemini prompt injection data leakage privacy attack automation LLM agents SpearBot attack personalization SPF DKIM DMARC MFA systematic literature review. true true false M. Zaluberšek Slovenski jezik Angleški jezik Diplomsko delo/naloga 2026-04-16 10:55:37 2026-07-07 10:18:01 2026-07-08 03:11:15 0000-00-00 00:00:00 2026 0 Ljubljana 2026 Ljubljana VI f., 30 str. 0000-00-00 Zaloznikova Objavljeno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 004.056(043.2) 283860739 VS_Zalubersek_Matic_2026.pdf VS_Zalubersek_Matic_2026.pdf 1 13EC2F9CEBA4C9C7C95C7EBA077B1C5D 701cf647026c9168491bf809391cec3dff23bf974e95e8f4a6a6c6885a19ff48 d97b0df4-4fb9-11f1-a431-00155d0001a1 https://dk.um.si/Dokument.php?lang=slv&id=206114 Fakulteta za varnostne vede 0 0 0