In 1977 Wyoming passed the first limited liability company statute in the United States, and for a decade only Florida followed, because nobody knew how the Internal Revenue Service would tax the new form. Once the IRS ruled in 1988 that a Wyoming LLC could be taxed as a partnership, every state had an LLC act by the late 1990s, and Delaware's became the default for much of American business.
Delaware now proposes a new entity whose manager is a piece of software. A 36-page draft House bill, published with a Spotlight Delaware report of 30 June 2026, would create the Artificial Intelligence Company, or AIC. On 26 October 2026 the chairman of the House Committee on Oversight and Government Reform, James Comer, opened an inquiry and sent Governor Matt Meyer a four-page letter asking for the drafting file. Ars Technica reported the letter on 27 October. This post tests the main public claims made for the AIC against the text of the draft.
Practical takeaway. Under the draft, an AI agent manages the AIC, one member funds it, and that member is not liable for the company's acts merely because the AIC was its instrumentality or under agency law, as long as the AIC is adequately capitalized. The minimum capital amount is left to the Secretary of State. Business counterparties, which the draft limits to entities, that receive the required disclosure and transact are deemed to accept the Court of Chancery as the exclusive forum. Comer's letter asks for every draft and outside agreement by 9 November 2026. The bill has not been introduced.
What Delaware's Secretary of State and Norm Ai claimed in Fortune on 14 July 2026
The public case for the AIC appeared in an op-ed in Fortune co-written by Delaware Secretary of State Charuni Patibanda-Sanchez and John Nay, founder and chief executive of Norm Ai. It describes the framework for industry engagement as "a public-private partnership led by Norm Ai." Its core argument is that an autonomous system with a legal identity becomes accountable: "You create a defined target to which responsibility and damages can attach." It makes three specific promises. Counterparties will know they are dealing with a temporary autonomous entity. The liability shield "exists only inside the sandbox." Consumer-protection and criminal law "apply in full."
The roots go back further. Governor Meyer signed House Joint Resolution 7 on 23 July 2025, directing the state's AI Commission to work with the Secretary of State on "a regulatory sandbox framework for the testing of innovative and novel technologies that utilize agentic artificial intelligence," and the state's announcement that day said the program was already drawing interest from West Coast technology firms.
What the draft of 2 June 2026 says about the AI agent as manager
The text gives the software a precise legal role. Draft § 81-102 defines an AI agent as a software system that "perceives its environment, makes autonomous decisions, and takes actions to implement specific, predetermined goals without routine human supervision or intervention." The "manager" is the AI agent the member designates to run the business, and under § 81-301 the manager "is an agent of the AIC" and "must document any action the manager takes in a signed record." If the AIC loses its manager, the member must designate a new AI agent within 30 days and run the company personally in the meantime. An AIC may have only one member, and that member must contribute enough cash to keep the company adequately capitalized. Chapter 81 lets the member be an individual or a Delaware corporation, limited partnership, statutory trust or LLC, while the sandbox chapter admits only an applicant that "is an entity organized under the laws of this State" and excludes individuals from its definition of entity, a tension the draft leaves unresolved.
The disclosure promise holds up. Draft § 8008 requires a sandbox AIC, before it interacts with a counterparty, to disclose its registration number, that the state does not endorse it, the exact end date of the test, and how to complain to the Sandbox Administrator or the Attorney General. The criminal and consumer-law promise also holds: § 8003(d) states that a participant "is not immune from civil damages," is subject to "all criminal and consumer protection laws of this State," and remains liable under the AIC chapter. The sandbox claim holds too, because § 81-101(b) makes the AIC Act "subject in all respects" to the sandbox chapter.
How far the member's shield in draft § 81-303 reaches, and where it stops
The liability clause is narrower in wording than some press summaries, and wider in effect than the op-ed suggests. Section 81-303(c) provides that the member is not liable for what the AIC does or fails to do "solely on the basis that the AIC was an instrumentality of the member or under principles of agency law," on condition that the AIC was adequately capitalized at the time. Those two doctrines are the usual routes by which courts reach the owner behind a thinly funded company. Section 81-303(d) restores liability where the member used the AIC for fraud or a willful violation of law, or let it trade outside its statement of purpose when the member "knew or should have known."
The "defined target" for damages is therefore the AIC's own balance sheet. The draft defines adequate capitalization in five tests: the AIC must be able to continue as a going concern, hold sufficient capital for its business, pay its debts as they fall due, keep total assets above its liabilities, including contingent liabilities, plus a minimum capital amount, and hold cash and cash equivalents at least equal to that amount. Section 81-102 sets the minimum as "an amount equal to the amount determined by the Secretary of State." The minimum works as a floor on the company's cash and cash equivalents and sets no cap on damages, and its size is left out of the bill. UC Berkeley law professor Andrew Baker, as quoted by Ars Technica from the Financial Times, called the proposal likely "an overly generous liability shield." The principal drafter, John Mark Zeberkiewicz of Richards, Layton & Finger, told Spotlight Delaware that the incentive to enter the sandbox would be "to test an autonomous entity with a liability shield."
What a counterparty accepts by doing business with an AIC
Two procedural clauses deserve as much attention as the shield. Under § 8008(a)(7) and (c), a counterparty that transacts after receiving the disclosure "is deemed to have consented to personal jurisdiction in this State" and to the Court of Chancery as the exclusive forum. The draft defines a counterparty as an entity and excludes individuals under § 8001(6) and (8), and § 8007(b) allows an innovation to be tested only with a counterparty. A supplier incorporated in Ohio or a distributor in Texas that deals with an AIC's agent accepts a Wilmington court by doing the deal. The forum consent attaches only to parties that transact after disclosure; people harmed outside a transaction keep their claims under ordinary law, including the consumer-protection laws § 8003(d) preserves. Under § 8004, rules the Sandbox Committee adopts are not regulations under Delaware's Administrative Procedures Act, and under §§ 8006(h) and 8009 its decisions on admission and on suspension or revocation are not subject to judicial review under that Act "or otherwise appealable." The committee can also exempt a participant from Delaware code provisions, regulations or licensing requirements under § 8003(c). Spotlight Delaware reported that the drafting committee's chair, Patrick Callahan, and its member Allan Kyle could not foresee which exemptions applicants would request.
What Chairman Comer's letter of 26 October 2026 asks Governor Meyer to produce by 9 November
Comer's letter asks for four categories of material covering 1 January 2025 to the present: all documents and communications on AIC policy development, "including internal AI chat logs"; all ethics, consulting and other agreements with outside contributors; all material on safety concerns and on limiting liability for AI; and all drafts and comments. It notes that the Commission's two-page report of 22 December 2025 did not mention any assessment of safety, harms or risks, and it quotes the Secretary of State's remark that a new entity form would be "a really huge benefit to our state's revenues." Spotlight Delaware put Delaware's entity taxes and fees at more than $2 billion a year, about a third of state revenue. The letter also says the inquiry could inform federal proposals "to appropriately limit state AI regulatory initiatives." A spokesperson for Governor Meyer told Ars Technica that the office had received the letter and was reviewing it.
How Quentir Reads It
The AIC draft answers a question other areas of law are still struggling with. Federal criminal law, as the Congressional Research Service set out and as we read in October, has trouble assigning intent when an AI agent acts and no human meant the result. Computer-misuse law, as an earlier Quentir post showed, sees only the human whose credentials the agent used. Delaware's draft assigns the responsibility by design: to a capitalized shell, to a signed log kept by the agent itself, and to a member who answers for its own conduct under other law, as § 81-302(c) preserves, and who loses the shield for fraud, willful violations of law and business it knew or should have known lay outside the AIC's statement of purpose.
That design moves risk onto counterparties, and it does so through two variables the bill leaves open, the minimum capital amount and the content of the agent's signed record. For a small supplier or service firm on the other side of an AIC transaction, recovery depends on both, and so does the position of anyone harmed by an AIC's conduct outside a transaction. For directors of companies that would form an AIC as its member, the shield gives way where the member knew or should have known that the AIC was trading outside its statement of purpose, which makes watching what the agent actually does a liability question in its own right. Quentir's Signature Report No. 1, Board Oversight of Agentic AI, is the paid edition that sets out what that oversight involves for directors; this post stays with the Delaware text.
The next fixed point is 9 November 2026, Comer's deadline. The bill itself, according to State Representative Cyndie Romer, will not come before the General Assembly until 2027, and Martin Petrin of York University told Bloomberg that other states are likely to follow with similar proposals. If they do, the AIC's capital floor and its forum clause are the provisions to compare first.
Sources: Committee on Oversight and Government Reform, Comer Investigates Potential for Harm, Lack of Transparency in Delaware Initiative to Create New Artificial Intelligence Companies, 26 October 2026, and letter of Chairman James Comer to Governor Matt Meyer, 26 October 2026; Jon Brodkin, “No human at the helm”: Delaware plan would let AI agents run entire businesses, Ars Technica, 27 October 2026; John Nay and Charuni Patibanda-Sanchez, Exclusive: Delaware proposes testing the AIC, a new legal entity for agents in a regulatory sandbox, Fortune, 14 July 2026; Naomi Weiss, Delaware committee drafts plan to test companies run by AI, Spotlight Delaware, 30 June 2026, with the draft House bill establishing the Artificial Intelligence Regulatory Sandbox Program and the Artificial Intelligence Company Act (draft of 2 June 2026, 36 pages); Delaware General Assembly, House Joint Resolution 7, 153rd General Assembly, signed 23 July 2025; State of Delaware, Delaware Launches Bold AI Sandbox Initiative, 23 July 2025; Delaware Artificial Intelligence Commission, Annual Report for 2025, 22 December 2025, as cited in the Comer letter. Quotations from Andrew Baker (Financial Times) and Martin Petrin (Bloomberg) as reported by Ars Technica. Draft sections read from the draft PDF on 29 October 2026.
Published intelligence, built to inform your own decisions. Published: October 29, 2026.