TY - GEN
T1 - A Systematic Analysis of AGI and ASI Existential Risk Scenarios
AU - Kumar, Yulia
AU - Li, Juan Jenny
AU - Kruger, Dov
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.
PY - 2026
Y1 - 2026
N2 - The rapid progress in Artificial Intelligence (AI) and statements by some leading figures in the industry suggest that Artificial General Intelligence (AGI) may be on the verge of becoming a reality. The timeline is highly uncertain, but once AGI emerges, it can be used to refine its own efficiency, potentially resulting in an explosive increase in capability known as a singularity. The potential for far-reaching effects demands that we consider the possible outcomes and risks. Artificial superintelligence (ASI) represents the next level of AI, at which it would surpass even the most gifted humans in cognitive competence. This paper critically examines the potential end-game implications of AGI and ASI systems, particularly the presence or absence of human-in-the-loop oversight. Building on recent advancements in large language models (LLMs) and multimodal AI, the study investigates how advanced AI might inadvertently or deliberately erode safeguards against weapons proliferation, exacerbate geopolitical tensions, and undermine global controls on high-risk technologies. Further, it offers a comprehensive taxonomy of existential risks associated with AI, including Artificial Narrow Intelligence (ANI), AGI, and ASI, culminating in the introduction of a pioneering multimodal nuclear weapon-related jailbreaking study, revealing critical vulnerabilities and potential misuse of next-generation AI systems.
AB - The rapid progress in Artificial Intelligence (AI) and statements by some leading figures in the industry suggest that Artificial General Intelligence (AGI) may be on the verge of becoming a reality. The timeline is highly uncertain, but once AGI emerges, it can be used to refine its own efficiency, potentially resulting in an explosive increase in capability known as a singularity. The potential for far-reaching effects demands that we consider the possible outcomes and risks. Artificial superintelligence (ASI) represents the next level of AI, at which it would surpass even the most gifted humans in cognitive competence. This paper critically examines the potential end-game implications of AGI and ASI systems, particularly the presence or absence of human-in-the-loop oversight. Building on recent advancements in large language models (LLMs) and multimodal AI, the study investigates how advanced AI might inadvertently or deliberately erode safeguards against weapons proliferation, exacerbate geopolitical tensions, and undermine global controls on high-risk technologies. Further, it offers a comprehensive taxonomy of existential risks associated with AI, including Artificial Narrow Intelligence (ANI), AGI, and ASI, culminating in the introduction of a pioneering multimodal nuclear weapon-related jailbreaking study, revealing critical vulnerabilities and potential misuse of next-generation AI systems.
KW - AGI
KW - ASI
KW - Artificial Narrow Intelligence (ANI)
KW - Existential Risk
KW - Human-in-the-Loop (HITL)
KW - Jailbreaking
KW - Multimodal AI
KW - Nuclear Weapons Proliferation
UR - https://www.scopus.com/pages/publications/105027553829
U2 - 10.1007/978-3-032-12313-8_2
DO - 10.1007/978-3-032-12313-8_2
M3 - Conference contribution
AN - SCOPUS:105027553829
SN - 9783032123121
T3 - Communications in Computer and Information Science
SP - 17
EP - 37
BT - AI Revolution
A2 - Arabnia, Hamid R.
A2 - Deligiannidis, Leonidas
A2 - Amirian, Soheyla
A2 - Ghareh Mohammadi, Farid
A2 - Shenavarmasouleh, Farzan
PB - Springer Science and Business Media Deutschland GmbH
T2 - International Conference on the AI Revolution: Research, Ethics, and Society, AIR-RES 2025
Y2 - 14 April 2025 through 16 April 2025
ER -