House Oversight and Government Reform Committee Chairman James Comer sent a letter on October 26 to Delaware Governor Matt Meyer asking for records about the state's plan to create a corporate form in which an artificial intelligence agent runs the business. The letter asks for documents and communications dating to January 1, 2025, including outside consulting agreements, draft legislation, risk assessments, and internal AI chat logs, with a response deadline of November 9.
The proposal, known as the Artificial Intelligence Company or AIC, would let an AI agent handle day-to-day operations with no human officer at the helm. A human member would fund the company and hold its economic interest, and would be shielded from liability except in limited circumstances. Delaware released a 36-page draft in June and expects to put the measure before the legislature in January 2027.
The letter opens a congressional inquiry into the plan. It argues that the AIC could protect AI companies and their investors from responsibility for harm, and that the process that produced the draft lacked transparency. The committee's findings, Comer wrote, could inform federal legislation meant to limit state AI rules. Delaware's office of the governor said it is reviewing the letter.
The letter asks for the paper trail
The document demand is unusually specific. Comer wants the risk assessments behind the proposal, the outside consulting agreements that shaped it, the draft bills and amendments, and communications between state officials and outside parties since the start of 2025. The request for internal AI chat logs suggests the committee wants to see how the draft was written, not only who reviewed it.
The letter also points to a December 22, 2025 report from the Delaware AI Commission. Comer says the report did not address safety risks, potential harms, risk assessments, or independent expert input, and that the public record does not show how the state weighed those questions before releasing the draft. His letter describes the AIC as a possible case of regulatory capture, in which the rules are written for the benefit of the companies they are meant to govern.
The committee is also asking about the state's relationship with Norm Ai, an AI compliance company whose chief executive, John Nay, announced the proposal with Secretary of State Charuni Patibanda-Sanchez in a Fortune opinion piece in July. The Secretary of State's office declined to answer questions about the working relationship, according to WHYY. The proposal's principal drafter is John Mark Zeberkiewicz, a lawyer at Richards, Layton & Finger who has represented Facebook and Meta in the past.
The request for AI chat logs is unusual in a corporate-law inquiry, and it cuts in two directions. It could show whether state officials used AI tools to draft the very statute meant to govern AI companies. It could also show how much of the proposal originated with Norm Ai rather than with the state. Either finding would be relevant to Comer's regulatory-capture argument, and both are hard to explain away once the logs are in the committee's hands.
The company would have no human at the helm
The AIC is a new kind of legal person. Under the draft, an AI agent would run the company's daily operations. The company could sign contracts, own and dispose of property, and sue and be sued in its own name. It could be formed for any lawful purpose other than banking, and it would operate inside a 30-month regulatory sandbox meant to let the state observe how the form behaves before deciding whether to make it permanent.
The human member would supply capital and receive profits, but would not manage the business. The draft requires that member to adequately capitalize the company, and it requires the company to keep an activity log of the AI's decisions. Those requirements are meant to create a record that regulators, courts, and counterparties can use when something goes wrong.
The form is not unlimited. The draft keeps banking out, and it subjects the company to existing law. The question the draft does not answer is how a company made of software and capital is supposed to satisfy obligations that assume a human decision-maker, such as the duty of care that officers and directors owe under existing corporate law. The AIC does not have officers or directors in the traditional sense, so the draft has to replace those duties with something else.
A sandbox is a familiar tool in financial regulation, where firms test products under supervision and the regulator collects data before deciding whether to permit them generally. The AIC draft borrows that structure and applies it to an entity form. The difference is that a financial sandbox typically has a defined consumer-protection floor and a regulator with authority to examine the firms inside it. The draft leaves many of those details to rules the state would write after enactment.
The liability shield is the contested part
The member's shield is the provision that has drawn the most criticism. Under the draft, the member is not liable for the company's obligations unless the company is undercapitalized, engages in fraud, or intentionally breaks the law. The AI itself cannot be sued or held liable in any conventional sense, so the shield effectively limits who can be held responsible when an AI-run company causes harm.
Andrew Baker, a law professor at the University of California, Berkeley, has called the shield excessively generous. David Hoffman of the University of Pennsylvania has described the AIC as a solution in search of a problem and warned that it risks outsourcing liability to a structure that cannot bear it. Christopher Bruner of the University of Georgia has said no broad plan he has seen addresses the fallout from a company governed by an AI agent.
Supporters see the same design as the point. Lawrence Cunningham of the University of Delaware has argued that ignoring AI-run companies would be irresponsible and that a sandbox lets the state learn from real cases instead of waiting for a crisis. The debate is less about whether the shield is generous than about what it is generous relative to: a legal system that currently has no way to register a company whose manager is a model.
Delaware's franchise business is the backdrop
Delaware earns more than $2 billion a year from corporate franchise fees and related revenue, about one-third of the state's income. The figure explains both the motivation behind the AIC and the sensitivity of the inquiry. The state has built its budget on being the default home for American companies, and it has competed for that business by offering predictable courts and a legislature that responds to corporate needs.
The AIC is an attempt to extend that franchise into a new market. If AI-run companies become common, Delaware wants the registration fees and the litigation that comes with them. The state's courts already handle a large share of corporate disputes, and its judges are used to applying familiar principles to new structures.
The risk is that the same business model produces a race toward the most permissive rules. Comer's letter frames the AIC as an example of a state competing for corporate charters by limiting liability, and it asks whether the public interest was considered alongside the revenue. Delaware officials have said the sandbox is a controlled experiment with reporting requirements, not a deregulation.
Congress is testing a limit on state AI rules
The letter's most consequential sentence may be the one about federal legislation. Comer wrote that the committee's findings could inform efforts to limit state AI regulatory initiatives. That is a signal that the inquiry is not only about Delaware's draft. It is part of a broader fight over whether states or the federal government set the rules for AI.
That fight is already running through Congress and the courts. States have moved ahead with AI laws on hiring, health insurance, and disclosure, and some members of Congress have argued for a federal framework that preempts them. A committee report on Delaware's process could become evidence in that debate, whether or not the AIC ever becomes law.
For Delaware, the letter creates a dilemma. Cooperating fully means handing Congress a record that may be used to argue for federal limits on state experimentation. Refusing or delaying means a public fight with a committee that can subpoena documents and hold hearings. The governor's office has said only that it is reviewing the request.
The draft is not law yet
The AIC still has to pass the legislature, and the January 2027 target is a goal, not a schedule. The 30-month sandbox would begin after enactment, and the state would have to decide what data to collect, how to audit the activity logs, and what would count as a failure serious enough to end the experiment early.
The legislature will also have to settle the questions the draft leaves open. What happens when an AI-run company commits a tort? Who pays the judgment if the member is shielded and the company's assets are gone? How does a counterparty verify that the AI has authority to sign? Those questions are not hypothetical; they are the daily work of corporate law, and the AIC has to answer them in statutory text.
If the legislature stalls or the committee's inquiry turns into a hearing, the draft could remain what it is now: a proposal with no legal effect. That outcome would still matter. The draft has already changed the terms of the debate over AI liability, and the questions it raised about who answers for an autonomous business will outlast the bill. Other states considering similar forms will have Delaware's experience, and Congress will have a record it can use.
For now, the proposal is a draft, a letter, and a deadline. The November 9 response will show how much Delaware is willing to disclose, and the committee's next step will show whether Congress wants a record or a fight. Either way, the question of who is responsible when software runs a company has moved from a state house in Dover to a committee room in Washington.
Primary sources
- House Committee on Oversight and Government Reform, letter and release on the Delaware AI company initiative, October 26, 2026. oversight.house.gov
- WHYY, report on the committee inquiry and the state's response, October 27, 2026. whyy.org
- Ars Technica, report on the AIC draft and its liability rules, October 27, 2026. arstechnica.com
- TechSpot, report on the proposal and the congressional response, October 28, 2026. techspot.com