ABSTRACT
Recent incidents such as the attack on Hugging Face by autonomous AI agents have sounded the alarm that a new breed of actor is entering our world. These agents are capable of imparting harm in a way that creates a gap in liability due to an inability to trace responsibility for their actions to an appropriate principal. This can be either because the principal is obscured or, more worryingly, because there is no principal at all. But this is not the first time that a class of non-human actors has gone about the world, sometimes in furtherance of performing a task for a principal and sometimes simply causing harm on its own. Using the analogy of canines, we classify AI agents both as they exist today and as they will exist in the impending future we foresee. We sort these agents on two axes: danger and domesticity. At either extreme is the Pomeranian, which is relatively safe and tied to an owner, and the Wolf, fully wild and dangerous. We look to history to see how people have sorted and dealt with the problems posed by canines, specifically medieval England and the Charter of the Forest.
The Charter of the Forest, and accompanying laws, sorted canines by domesticity and dangerousness and had a remedy for each. We propose a New Charter of the AI Forest that does the same. For domesticated AI agents, we propose an identification system, a digital collar, that provides information about an agent’s capabilities and responsible principal. For agents that have no identifiable principal, we propose the creation of a system of incorporation that gives such agents an identity. If an incorporated agent is to have access to the open network, it must have some kind of assets or insurance in order to address liability that may arise from its actions. Given network access restrictions, we believe that wild agents may choose this path of their own volition. In essence, this would be an act of self-domestication and help bring order and limit uncompensable harm in the agentic age.
Boddy, Seán and Darmon, Nathan and Ferrer, Lucas and Fink, Brigitte and Joseph, Joshua and Levine, Jacob and Weinstock, Jordi and Zittrain, Jonathan, Charter of the AI Forest: Tort Accountability and Legal Bodies for Autonomous AI Agents (October 15, 2026). Forthcoming in the Harvard Journal of Law & Technology, Symposium Edition, volume 41.3 (2027).
Leave a Reply