Do AI agents have free will, really?

Hauhio, I., Buder-Gröndahl, T. & Wallenberg (2026)

Martela (AI and Ethics 5: 4389–4400, 2025) claims that contemporary AI agents based on large language models (LLMs) possess a form of free will, passing the so-called List’s Agency Test. Martela presents two examples of such LLMs, a video game playing software Voyager and a fictional autonomous drone Spitenik. In this paper we argue that neither of these examples pass List’s Agency Test. Instead, Voyager’s and Spitenik’s behavior can be exhaustively explained computationally. Furthermore, to the extent such a computational account is available, the assignment of intentional contents can be seen as merely “glossing” the system’s internal computational states, in the sense elaborated by Egan. Intentional considerations are, thus, not explanatorily mandatory as required by List’s Agency Test. Since List’s conditions for intentionality are not met, Martela’s arguments for free will in AI agents fail.

Hauhio, I., Buder-Gröndahl, T. & Wallenberg, AM. Do AI agents have free will, really?. AI Ethics 6, 331 (2026). https://doi.org/10.1007/s43681-026-01125-z