Abstract The increasing availability of locally executed artificial intelligence systems enables individuals to deploy personalised, multi-agent AI environments outside institutional control. While such systems offer privacy and autonomy advantages, they also shift ethical, legal, and governance responsibilities from organisations to individual users. Current AI ethics frameworks largely address institutional deployments and provide limited guidance for personal AI systems. This paper presents the design and evaluation of Tibor, a lightweight governance agent embedded in a personalised multi-agent AI environment. Tibor introduces ethical oversight through structured metadata that influences anonymisation, information routing, retrieval-augmented generation, and output review, without restricting user autonomy. Using a design–science methodology supported by longitudinal reflective observation, the study examines how embedded governance affects privacy awareness, ethical decision-making, academic integrity, and safety-related behaviour during real-world use. The findings indicate that simple, user-facing governance mechanisms can meaningfully influence AI-supported workflows by prompting responsible data handling, flagging ethically sensitive content, and supporting informed decision-making. The governance functions operate transparently and with minimal operational overhead, illustrating the feasibility of ethics-by-design approaches at the individual level. The paper contributes a practical model for embedding ethical governance into personal AI systems and highlights the need to extend AI ethics discourse beyond organisational settings to address emerging forms of user-managed AI.
Gyula Szabó (Mon,) studied this question.