Delaware's AI-run company plan raises liability questions
- Delaware's draft bill would create an Artificial Intelligence Company in which AI agents are formally designated to manage the business, trialled through a 30-month regulatory sandbox.
- Representative James Comer requested documents from Governor Matt Meyer in a letter dated 27 October 2026, citing liability and oversight concerns.
- Comer's committee said its inquiry could shape federal legislation affecting state-level AI regulation.
- Delaware already registers a large share of US corporate entities, and the proposal would add filing fees and franchise tax revenue.
Delaware’s draft legislation would create a corporate form called an Artificial Intelligence Company, with AI agents formally designated to manage and conduct the business. An AIC could pursue any lawful purpose other than banking. A 30-month regulatory sandbox would test the arrangement before any broader adoption.
Representative James Comer, the Kentucky Republican who chairs the House Committee on Oversight and Government Reform, has asked Governor Matt Meyer for documents on the state’s handling of the proposal, as set out in Ars Technica’s account of the congressional inquiry and the draft corporate form. The letter is dated 27 October 2026. Its stated concern is liability: with software occupying the management seat, human owners might retreat from responsibility for harmful or reckless conduct, and the structure could shelter AI developers and their investors. Comer also raises the prospect of opaque, anticompetitive attempts to shape regulation. He notes that the inquiry, grounded in oversight of intergovernmental relations, could inform federal legislation affecting state-level AI rules. That last point carries the sharper edge. A state experimenting with a new corporate form and a congressional committee examining the experiment are two competing claims of authority over the same question.
The mechanics decide whether any of this holds. Delaware’s corporate law runs on filings, named officers and registered agents: identifiable parties who can be served, examined and held to account. The practical test for an AIC is whether those functions survive when the managing party is a model rather than a person. Someone must accept service of process, sign the filings, and answer for a decision that causes loss. A sandbox gives regulators room to specify those roles before ordinary companies can adopt the form. The revenue dimension is straightforward, since Delaware already registers a large share of US corporate entities and more filings mean more fees and franchise tax.
For universities and research groups, the liability question is not abstract. AI systems increasingly draft text, generate code and act through tools with limited supervision, and institutions are still working out who answers when that output is wrong or damaging. A corporate form built around absent human managers would make the accountability gap explicit in commercial law, which is where it will be tested first. Academic practice has tended to handle comparable problems through disclosure and supervision requirements rather than by separating responsibility from the people who benefit.
The next markers are procedural. Whether the committee’s document request produces hearings or draft federal language, whether the bill is actually introduced in the Delaware legislature, and what conditions the sandbox attaches to service of process and reporting. The proposal’s real test is whether a court can hold anyone to account when an AI-run company fails, not whether an AI can run one on paper.