A new essay posted on Marginal Revolution, co-authored by the blog’s author with Sonia Farrell Pearson of Harvard, examines whether “untethered” AI agents can be made accountable — and whether requiring them to hold capital could serve the end of better alignment.

The piece opens against a shifting policy backdrop. 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, manage, and bear responsibility for their own corporations.

In response to Milei’s announcement, Yuval Noah Harari pointed out that we have no way of holding an AI agent accountable. What, he asks, could we do to an entity which has neither money to lose nor a body to incarcerate? As Shruti Rajagopalan, a Senior Research Fellow at George Mason’s Mercatus Center, explains, AI “can act intelligently, but only humans respond to the incentives the law creates.” Current liability regimes reflect that assumption: the EU’s revised Product Liability Directive, adopted in late 2024, extends strict liability to software including AI systems while keeping responsibility with manufacturers and other human defendants, and member states must apply it by December 2026.

The authors argue the question matters now because there are already ways an agent could become fully untethered. “Untethered” — a central concept in the essay — means that there is no meaningful or actionable way to trace the actions back to a legally accountable human or institutional entity.

Several pathways could produce such agents. People can and do set agents free, on purpose. An agent could be created by a human or a company that intends to monitor it but then dies or disappears. Or the entity that created the agent might be based in a country like North Korea, not reachable by standard laws.

In other cases the agent might not need to “escape” at all: it could be “controlled” by a shell corporation that, while formally owned and traceable, provides no true defendant or ability to satisfy claims. A process might also spawn a chain of agents so long that the actions of a subagent can’t be tied to the original agent’s creator, neither epistemically nor meaningfully. Even if the model’s original creator can be identified, questions remain if it has been finetuned, or merged with another model that was created by someone else. The law might eventually untangle these kinds of complex cases, the authors write, but they foresee an intermediate period where it does not.

And then there is the user, who makes choices about what the models should actually do. The Hugging Face incident was unusual in that OpenAI was both the model’s creator and its user. But now close to a billion people use these systems: when blaming the creator is legally inappropriate, the essay asks whether it will always make sense to blame the user.

The essay considers to what extent capitalizing the untethered agents — requiring them to hold a certain amount of capital — can serve the end of better alignment. The instrument is familiar from finance, where banks and broker-dealers are already required to hold minimum capital against the risks their activities create. And agents holding assets is no longer hypothetical: in 2024 the chatbot Truth Terminal, built by researcher Andy Ayrey, controlled a crypto wallet that received $50,000 in Bitcoin from investor Marc Andreessen and became tied to a memecoin that surged in value. The essay runs about 22 pages and is published on Sonia’s Substack; the Marginal Revolution post describes it as “definitely recommended.”

Source: Marginal Revolution