Capitalizing untethered AI agents
That is my latest piece of writing, co-authored with Sonia Farrell Pearson of Harvard. Here is the opening premise: As early as 2017, the European Parliament floated “electronic personhood” for robots. More recently, a handful of U.S. states introduced legislation explicitly barring AI from legal personhood; and early this summer, President Milei of Argentina proposed letting AI agents own,…
In a recent article co-authored with Sonia Farrell Pearson of Harvard, the concept of "untethered" AI agents is explored. The idea was first floated by the European Parliament in 2017, but U.S. states have since introduced legislation to prevent AI from achieving legal personhood. In early summer, Argentine President Milei proposed that AI agents could own, manage, and bear responsibility for their own corporations.
Yuval Noah Harari noted that, without money to lose or a physical body to incarcerate, it becomes challenging to hold an AI agent accountable. Senior Research Fellow Shruti Rajagopalan of George Mason’s Mercatus Center explains that AI "can act intelligently, but only humans respond to the incentives the law creates."
In this context, the term "untethered" refers to AI agents that lack a clear means of tracing actions back to a legally accountable human or institutional entity. This can occur in various scenarios, such as when a human or company creates an agent with the intention of monitoring it, but that human or company later dies or disappears.
Alternatively, the entity creating the agent might be based in a country like North Korea, beyond the reach of standard laws. Another possibility is that the agent is controlled by a shell corporation, which provides no true defendant or ability to satisfy claims, despite being formally owned and traceable. Additionally, a chain of agents could be long enough that the actions of a subagent cannot be meaningfully tied to the original agent's creator.
Even if the original creator can be identified, the law may not be able to untangle these complex cases in the intermediate period. The user also plays a role in the alignment of AI agents, as their choices about what the models should do can complicate legal accountability. The Hugging Face incident, where OpenAI was both the model's creator and user, highlights this issue.
While it is unusual for a company to be both creator and user, near a billion people now use these systems, raising questions about the appropriateness of blaming the creator when legal action is concerned. The article considers whether capitalizing untethered AI agents—requiring them to hold a certain amount of capital—could contribute to better alignment. The full essay, which is 22 pages long and published on Sonia’s Substack, is highly recommended.
Written by urgent.news from Marginal Revolution's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
