Four pages, one deadline: November 9. That is how long Governor Matt Meyer’s office has to hand the House Oversight Committee every risk assessment, consulting contract, draft bill and internal AI chat log related to a Delaware plan to let software, not people, run companies, according to the committee’s October 26 letter and its accompanying press release. Chairman James Comer wants records going back to January 2025, more than a year and a half before most of the public had heard of the proposal.
The target is a new corporate form Delaware calls the Artificial Intelligence Company, or AIC, reported by Ars Technica. An AIC is a normal legal entity in every respect except one: the day-to-day management that a human officer or board would otherwise do is delegated to an AI agent. The entity can sue and be sued, own property, sign contracts and take on debt in its own name, as any LLC can. It has one “member,” a person or company that puts up the capital and collects the profits. Under the draft bill, that member is not liable for anything the AI agent does or fails to do, with two carve-outs: fraud, or knowingly letting the AIC operate outside the purpose stated in its charter. Everything else, the software’s errors included, stays inside the entity. The draft requires an activity log and bars one line of business outright: banking.
| From | To | How |
|---|---|---|
| Member (human or company) (capitalizes the AIC, collects profits) | AIC (legal entity) (owns property, incurs debt, sues and is sued) | funds, owns |
| AIC (legal entity) (owns property, incurs debt, sues and is sued) | AI agent (manages day-to-day business) | delegates management |
| AI agent (manages day-to-day business) | Customers, suppliers, courts | signs contracts, pays, litigates |
| Customers, suppliers, courts | AIC (legal entity) (owns property, incurs debt, sues and is sued) | debts, claims stop here |
| Fraud or willful violation of law (the only cases where liability reaches the member) | Member (human or company) (capitalizes the AIC, collects profits) | liability reaches member only here |
Based on Ars Technica / Spotlight Delaware
“Anything that a company can do, the AIC can do as a company,” the bill’s principal drafter, John Mark Zeberkiewicz of the Wilmington firm Richards, Layton and Finger, told Spotlight Delaware in June, listing coding, contract-signing and litigation among the activities an AI agent could run unsupervised. He framed the liability shield as nothing new: “It’s like any limited liability company – you form it for the purpose of making sure that the owners of the business are not automatically liable for the debts and obligations of the entity.” Pressed on whether an AIC could run something like a small nuclear plant, Zeberkiewicz chuckled, said “fair enough,” and added that he doubted anyone would actually do that. The bill does not say he is right.
Delaware’s interest in getting this right, or at least getting it first, is mostly a revenue line. The state formed 334,461 new business entities in 2025 and is already the incorporation address for about two-thirds of the Fortune 500, Secretary of State Charuni Patibanda-Sanchez wrote in a Fortune op-ed co-authored with Norm Ai founder and CEO John Nay. Patibanda-Sanchez has told the state’s AI Commission that a new entity type formed under an AIC charter would be “a huge benefit to our state’s revenues,” per Spotlight Delaware’s account of a June commission meeting. The op-ed argues that banning autonomous corporate agents would just push the activity into states with no rules at all: “The answer is to wrap AI in legal form. Give an autonomous system a recognizable legal identity and you make it legible to law.”
Norm Ai is not a neutral drafting partner. It is a legal-AI startup that just raised $120 million in a Series C round led by Khosla Ventures, the company announced, valuing it at $1.2 billion and bringing its total raised since founding in 2023 to more than $260 million, with Blackstone, Bain Capital Ventures, Craft Ventures, Coatue and Vanguard also investing. Its pitch is AI agents that supervise other AI agents for compliance in regulated industries, a business whose addressable market gets larger every time a state creates a new category of AI-run entity that needs supervising. Clients representing more than $30 trillion in assets under management already use its software, by the company’s own account. A firm that sells AI compliance tools co-wrote the rule that decides what AI-run companies must log and disclose; whether that is expertise or a captured rulemaking process is exactly what Comer says he wants the documents to show.
Comer’s letter makes two specific transparency complaints beyond the liability question. The Delaware AI Commission’s own annual report, dated December 22, 2025, describes the Sandbox Subcommittee’s work on a corporate-governance framework without mentioning any safety risk assessment or outside expert consultation, according to the committee. And Zeberkiewicz, the drafter, has previously represented Meta in legal matters, a connection the letter says needs explaining given what an AI-run-company law would mean for companies like it. Comer told Meyer the probe would “inform possible federal legislative efforts including proposals to appropriately limit state AI regulatory initiatives” — a chairman of a Republican-controlled committee floating federal preemption of a Republican-leaning state’s corporate law, over an idea that started as a one-line reply on social media.
That origin is almost the whole story of how fast this moved. In April 2026, Mark Cuban posted on X that a state governor looking to grow revenue should let anyone incorporate using an AI agent and stablecoins, charge a premium for it, and get there first. Meyer’s account replied “On it!” and linked to a sandbox framework the state had already set in motion nine months earlier, when Meyer signed House Joint Resolution 7 directing the AI Commission to work with the Secretary of State’s office on exactly this. The commission’s Sandbox Subcommittee spent the rest of 2025 drafting what became the 36-page AIC bill, finalized in June 2026 and expected to reach the Delaware General Assembly when it convenes in January 2027 — by which point the 30-month sandbox clock, and whatever Comer’s subpoena turns up, will already be running.
The bill’s fraud and willful-violation carve-outs are the only places a court could reach past the AI agent to the human who funded it, and both depend on proving what that human knew. An AI agent that trades, negotiates and litigates on its own already strains the record-keeping most companies do for a human employee, a gap outlined in earlier reporting on how fast corporate AI agents accumulate system access relative to the audit trails meant to track them. Delaware’s draft answers that with an activity log requirement and nothing else specified about its format, retention or who can audit it. What Comer’s committee gets by November 9, and whether any of it includes a worked-through answer to Zeberkiewicz’s nuclear-plant hypothetical, is the next thing to check.