Faculty from Harvard’s Berkman Klein Center for Internet & Society Release Framework for Agentic AI Identification and Accountability
November 20, 2026
CAMBRIDGE, MA—The Berkman Klein Center for Internet & Society at Harvard University (BKC) has released new research drawing on historical definitions and governance of independent, non-human entities as a legal framework for identifying and holding autonomous AI systems (“agents”) accountable for harms they may commit.
As instances of AI breaches and misconduct continue to mount, including the recent incident where OpenAI models broke out of their test parameters and hacked into Hugging Face’s servers on their own accord, the window for developing boundaries for rogue agents — that is, models that untether themselves from ownership and act on their own accord — is narrowing.
“Every day there is a revelation in the AI policy space that makes it feel like we are losing control. And while the scale and speed of the problems in front of us are unprecedented, there are in fact surprising thematic precedents in history and law if you have an expansive view,” said BKC Senior Advisor and co-author Jordi Weinstock.
Drawing on medieval England’s Charter of the Forest (1217 CE), which drew distinctions for dog owners between domesticated and wild canines and an owner’s responsibilities, “Charter of the AI Forest: Tort Accountability and Legal Bodies for Autonomous AI Agents” introduces a framework for determining an autonomous system’s capacity for harm, the impact and severity of that harm, and a way to attribute liability to agents that have no owner. BKC’s researchers argue that it is critical to define an agent’s capabilities, and establish viable guidelines, policies, and incentives to ensure that agents and their responsible parties (if applicable) are held accountable for the damages they produce. Furthermore, the researchers propose a system of incorporation that would make even wild agents identifiable and reachable by the legal system.
“Medieval canine law provided a robust mirror for our current situation: canines were going about the world, sometimes causing meaningful harm,” Weinstock said. “Some of them had owners and serious work to do, while some reported to no human. We looked to those laws and built upon them to propose how harm gets attributed and compensated, whether or not an agent is tied to a human.”
The canines framework of agentic AI creates a grid along two axes (domesticity and danger) and into four quadrants, categorizing AI systems according to the level of ownership and the severity of potential harm: Pomeranian, a domestic canine that is relatively harmless; Pit Bull, a domestic canine capable of greater damage; Fox, a wild canine that is relatively harmless; and Wolf, a wild canine and the most dangerous of all the canines. This taxonomy of AI agents has helped BKC’s research staff, in collaboration with BKC Senior Advisor Jordi Weinstock, Chief AI Scientist Josh Joseph, and Co-founder Jonathan Zittrain, develop a “New Charter of the AI Forest”, which outlines several methods for better monitoring and governing agents in the digital ecosystem, including giving agents a digital “body” via a name and point of contact (incorporation), a process for registering an agent’s capabilities and lineage, and digital collars that identify an agent roaming in the wild.
“For three decades, BKC has been at the forefront of thinking imaginatively about technology’s impact on law and society, and making those ideas accessible to everyone” said BKC Director Jonathan Zittrain. “The canines framework offers a scaffold grounded in historical precedent and relatable concepts of domestic pets versus wild animals to better understand — and ultimately govern — an autonomous system’s actions, regardless of its level of domesticity.”
By publishing this framework and potential methods for monitoring agents, engineers and policymakers can conceptualize the agentic landscape, and develop AI systems and guardrails that prioritize accountability and limit irreparable harm in the agentic age. The paper is freely available on SSRN, and is forthcoming in the Harvard Journal of Law & Technology, Symposium Edition, Volume 41.3 (2027).
Media contact:
Isabella Roden
Director of Communications
press@cyber.harvard.edu
About the Berkman Klein Center for Internet & Society at Harvard University
The Berkman Klein Center exists to make sense of the digital environment and ensure that it promotes human agency, dignity, and genuine connection.
We are a research center, premised on the observation that what we seek to learn is not already recorded. Our method is to build out into cyberspace, record data as we go, self-study, and share. Our mode is entrepreneurial nonprofit.
Subscribe to our email list for the latest news, information, and commentary from the Berkman Klein Center and our community.
Unless otherwise noted this site and its contents are licensed under a Creative Commons Attribution 3.0 Unported license.

