Delaware’s draft would let an AI run the company.
By Ellis Ward | November 8, 2026
A Company With No Human at the Helm
Delaware has been drafting a plan to let an AI agent run a legal company. No human executive is required. The proposal is 36 pages long and not yet law. But it has already drawn a congressional inquiry, a Fortune op-ed from the state's own secretary of state, and a federal document deadline of November 9.
The idea is called the Artificial Intelligence Company, or AIC. Under the draft, the entity can hold property, take on obligations, and sue or be sued in its own name. An AI agent manages its affairs and conducts business for any lawful purpose except banking. A single human member funds it and collects the profits. That member generally does not answer for what the AIC does.
That last sentence is where the trouble starts.
The Mechanics: What an AIC Actually Is
Governor Matt Meyer signed a resolution in July 2025 establishing the Delaware Artificial Intelligence Commission. Its mandate was to develop the AIC framework. By June 2026, a 36-page draft bill had been created. Corporate lawyer John Mark Zeberkiewicz is its principal drafter.
The structure borrows from limited liability concepts but pushes them further. A traditional LLC shields its members from personal liability for company debts. An AIC does the same thing, but the operating party is software. The AI agent signs contracts, spends money, and runs day-to-day operations. The human member sets up the entity, funds it, and waits for returns.
The draft lists four exceptions to the liability shield. A member can be held responsible for fraud, undercapitalization, willful unlawful conduct, or negligent operation outside the AIC's stated purpose. Outside those four situations, the human behind the AIC is generally protected.
The 30-Month Sandbox
Alongside the AIC framework, the draft includes a 30-month regulatory sandbox. AI-managed entities operating under it would test the structure before full rules apply. Regulators and participants would have time to watch the mechanics work, or fail, before the framework locks in.
The timeline matters. The draft bill was described in June 2026 as likely to be introduced in the legislature as early as January 2027. If it passes, the sandbox clock starts immediately. Real AI-managed companies could operate under limited oversight for two and a half years before full regulatory exposure kicks in.
That is a long window for a framework with no precedent.
The Accountability Gap
The liability shield is a feature of the draft, not a flaw in it. Delaware's implicit argument is that separating human liability from AI conduct is what makes the structure useful. Investors will not fund a vehicle they can be personally sued over. The four exceptions are designed to catch genuine bad actors, not ordinary business risk.
Critics see a gap. When a human executive makes a bad call, there is a chain of accountability. The executive, the board, and the company all carry some exposure. When an AI agent makes the same call, the chain is shorter. The company exists. The human member is generally protected. The AI has no assets of its own.
In July 2026, Delaware Secretary of State Charuni Patibanda-Sanchez and John Nay, founder of Norm Ai, published an op-ed in Fortune that addressed this directly. Their argument was that legal recognition of an autonomous system makes its conduct visible. It attaches responsibility and damages to a defined entity. Without legal personhood, harms caused by an AI-managed operation are harder to trace and harder to remedy.
That argument flips the accountability concern. The question is not only who answers when an AIC causes harm. It is also whether, without an AIC structure, anyone answers at all.
The Contract Problem
Consider how the mechanics play out in practice. An AI agent operating under an AIC structure signs a supply contract. The supplier delivers. The AIC does not pay. The supplier can sue the AIC, which holds property and can be a named defendant. The human member generally does not pay from personal funds unless one of the four exceptions applies.
That mirrors how an LLC works today. A supplier stiffed by an LLC sues the LLC, not the founding member, unless fraud or undercapitalization is involved. The AIC framework extends that logic to an operating party that is software.
The harder case involves harm that a court judgment cannot fully compensate. If an AIC-managed operation causes physical damage, the AIC is the defendant. Its assets pay the claim. If those assets run short, the member is likely protected. The human who set up the entity, funded it, and profited from it may face no further exposure.
The four exceptions are meant to reach extreme cases. Fraud requires intent. Willful unlawful conduct requires a knowing violation. Undercapitalization requires that the member funded the entity inadequately from the start. Negligent operation outside the stated purpose requires that the AIC acted beyond what it was set up to do.
A member who funds an AIC adequately and watches the AI make a series of poor but lawful decisions inside the stated purpose likely stays protected. That is the gap critics point at.
Congress Demands Answers
House Oversight Committee Chairman James Comer sent a letter to Governor Meyer on October 26, 2026. He requested documents on how the AIC proposal was developed and how AI safety was considered in the drafting process. Documents are due by November 9, 2026.
The letter did not stop at a document request. Comer warned that the AIC structure could allow human owners to avoid responsibility for reckless and harmful actions by software they appoint to run a business. He stated that the committee's review could inform federal legislation limiting state AI initiatives.
That second point is more than a procedural note. It signals that Congress might act to restrict what states can do with AI governance entirely. Delaware's framework would stop being a state experiment and become a flashpoint for where federal authority ends and state corporate law begins.
A spokesperson for Governor Meyer's office said the governor's office received Comer's letter and was reviewing it. No misconduct has been established. The proposal is not law.
The Federal-State Fault Line
Delaware has spent decades as the dominant state for corporate formation. The AIC proposal fits that history. Delaware writes the rules that companies want to operate under. Other states follow.
Federal interest in the AIC framework represents a different kind of pressure. If Congress concludes that state AI governance creates unacceptable liability gaps, it could preempt state action. That would shift control over AI entity law from state legislatures to federal agencies or Congress itself.
That fight has not happened yet. Comer's letter is an inquiry, not legislation. The November 9 deadline is a data-gathering step. But the warning embedded in the letter, that the committee's findings could inform federal legislation, sets up a confrontation between Delaware's corporate law tradition and federal ambitions over AI oversight.
Delaware's response to that letter will tell a lot about how the state intends to defend the framework.
Who Benefits, Who Bears the Risk
The AIC structure as drafted sends the upside to the human member and absorbs the downside into the entity, or in limited cases, back to the member through the four exceptions. That is not unusual in itself. Limited liability is the foundation of modern corporate law. LLCs exist to let investors take risks without staking their personal assets on every outcome.
What differs with an AIC is the identity of the decision-maker. A human CEO who runs a company into the ground can face a breach of fiduciary duty claim, an investigation for fraud, or public accountability. An AI agent that makes the same decisions has no professional reputation at stake, no personal liability to threaten, and no intent to prove or disprove.
The four exceptions try to reach the human behind the AIC in extreme cases. The thresholds are high. Ordinary negligence in a lawful operation where the AI simply made a bad call likely falls outside all four of them.
Visibility vs. Responsibility
The Fortune op-ed by Patibanda-Sanchez and Nay makes a point worth examining carefully. Legal recognition creates a target for lawsuits. Without a legal structure, harms caused by autonomous systems are harder to attribute and harder to compensate. The AIC, in that reading, does not shield bad actors. It makes AI conduct legible under existing law.
The counterargument, implied in Comer's letter, is that the target is too thin. An entity with limited assets, no human decision-maker, and a member shielded by a liability wall is not a meaningful place to send a damages claim. It is a placeholder. The real question is whether the law should require more from the humans who create, fund, and profit from AI-managed companies.
The current draft does not answer that question cleanly.
What Happens Next
The November 9 deadline for Governor Meyer's office is the immediate pressure point. If the governor's office cooperates fully, the committee gets its documents and the review proceeds on its own timeline. If the response is incomplete, Comer has suggested federal legislation could follow.
The AIC bill is expected to be introduced in the Delaware legislature as early as January 2027. That is a prediction, not a commitment. The federal inquiry adds a variable. Drafters may revise the liability provisions in response to congressional pressure. The sandbox structure may be adjusted. Or the bill moves forward unchanged and tests whether Delaware's corporate law tradition can hold ground.
The 30-month sandbox is designed to generate real data before full rules apply. If the AIC framework passes and the sandbox runs, Congress and regulators will have actual cases to examine, not just a 36-page draft. That changes the debate. It shifts the question from what might happen to what has already happened.
No AI-managed company exists under this framework yet. The AIC is still a proposal, working its way toward a legislative calendar that may or may not hold. But the debate it has triggered, over liability, federal power, and what it means for software to run a legal entity, is already moving faster than the legislation.
