Ethical Issues With LLM-Based AI Agents : A Systematic Literature Review
Pesä, Paavo (2025)
2025
Sähkötekniikan DI-ohjelma - Master's Programme in Electrical Engineering
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
Hyväksymispäivämäärä
2025-07-28
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202507287837
https://urn.fi/URN:NBN:fi:tuni-202507287837
Tiivistelmä
This thesis presents a systematic literature review (SLR) on the ethical issues associated with large language model (LLM)-based artificial intelligence (AI) agents - systems that combine advanced language understanding with autonomous, interactive, and goal-directed capabilities. The study addresses a research gap where existing ethical analyses have focused primarily on standalone LLMs or general AI systems, overlooking the unique challenges introduced by agen-tic architectures. In addition to identifying ethical challenges, the review also examines how AI agents are defined and conceptualized across the literature. A total of 64 peer-reviewed studies published between 2021 and 2025 were reviewed, leading to the identification of 22 recurring ethical issues.
These issues are categorized using the Ethics Guidelines for Trustworthy AI, published by the European Commission’s High-Level Expert Group on AI. The guidelines outline seven key themes for trustworthy and ethical AI: human agency and oversight; technical robustness and safety; privacy and data governance; transparency; diversity, non-discrimination, and fairness; societal and environmental well-being; and accountability. The findings show that human agen-cy and oversight, transparency, and accountability are the most frequently addressed themes amongst AI agents, reflecting a research emphasis on user-facing concerns such as control, explainability, and responsibility. In contrast, broader systemic issues such as environmental sustainability and societal impact remain underexplored.
The review highlights the lack of consensus on the definition of AI agents, the dominance of technical solutions over normative discourse, and the limited operationalization of ethical prin-ciples. By synthesizing key challenges and research gaps, this thesis contributes to the devel-opment of more robust ethical frameworks for LLM-based AI agents and provides guidance for future research to bridge the gap between abstract ethical guidelines and their practical imple-mentation.
These issues are categorized using the Ethics Guidelines for Trustworthy AI, published by the European Commission’s High-Level Expert Group on AI. The guidelines outline seven key themes for trustworthy and ethical AI: human agency and oversight; technical robustness and safety; privacy and data governance; transparency; diversity, non-discrimination, and fairness; societal and environmental well-being; and accountability. The findings show that human agen-cy and oversight, transparency, and accountability are the most frequently addressed themes amongst AI agents, reflecting a research emphasis on user-facing concerns such as control, explainability, and responsibility. In contrast, broader systemic issues such as environmental sustainability and societal impact remain underexplored.
The review highlights the lack of consensus on the definition of AI agents, the dominance of technical solutions over normative discourse, and the limited operationalization of ethical prin-ciples. By synthesizing key challenges and research gaps, this thesis contributes to the devel-opment of more robust ethical frameworks for LLM-based AI agents and provides guidance for future research to bridge the gap between abstract ethical guidelines and their practical imple-mentation.