What Delaware is proposing
In a July 14, 2026 piece for Fortune, Delaware Secretary of State Charuni Patibanda-Sanchez and Norm Ai founder John Nay laid out a new corporate form they call the Artificial Intelligence Company, or AIC. An AIC would be a separate legal entity managed by an AI agent rather than a person. In the authors' words, it could sue and be sued in its own name, hold and dispose of property, and incur obligations. The idea gives an autonomous agent a recognized legal home, so that when it negotiates a contract or takes on a liability, there is a defined entity standing behind the action rather than an ambiguous chain of software and operators.
Delaware matters here because most large U.S. companies incorporate there, and its corporate law tends to set national precedent. The AIC concept extends that franchise to a class of business activity that agentic AI is already starting to perform, from coding services to contract negotiation. Rather than waiting for courts to sort out liability after the fact, the proposal tries to define accountability up front. For technology leaders watching agents move from assistants to actors that transact, the significance is that a major jurisdiction is now sketching the legal container those actors would operate inside.
How the sandbox would work
The AIC would exist only inside a regulatory sandbox, and admission would run through a committee that includes the Delaware Secretary of State, the Attorney General, the Chief Justice of the Delaware Supreme Court, the chair of the state AI Commission, plus outside attorneys and technologists. Participants would have to maintain activity logs and disclose their status to anyone they deal with, including that the entity is temporary, carries no state endorsement, has a defined test end date, and offers a complaint procedure. Officials would retain the power to suspend an authorization, revoke a license, and ask a court to dissolve an AIC that misbehaves.
The liability design is deliberately narrow. A single human or corporate member sits behind the AIC and is shielded from its debts, except where there is a failure to capitalize the entity, fraud, or a willful violation of law. Consumer protection statutes and criminal law apply in full, and banking is carved out of the program entirely. The whole scheme would sunset after 30 months, giving the legislature a defined window to judge results before deciding whether to make anything permanent. The structure reads as a controlled experiment with clear off-switches rather than an open grant of autonomy.
Who is behind it and why it is framed this way
The framework is being developed as a public-private partnership led by Norm Ai, a company that builds AI for regulatory compliance, alongside the Secretary of State's office. That pairing shapes the pitch. The authors present the AIC as a way to bring autonomous commerce inside a predictable American legal order before it scales past the reach of existing entity law. Their argument is blunt: agents are already doing business, and the question is whether that activity happens under supervision or in the gaps.
The offshore-migration case is central. Nay and Patibanda-Sanchez warn that if the United States does not offer a careful, accountable structure, the activity will migrate offshore to jurisdictions with looser oversight. That framing will resonate with policymakers who watched crypto and other fast-moving sectors route around domestic rules. It also positions Delaware to compete for a category of incorporation that does not yet exist, much as it competed for conventional company formations over the past century. The commercial and regulatory motives here are aligned rather than opposed.
How it fits the broader agent-governance moment
The AIC proposal arrives as governments worldwide grapple with autonomous agents. China's implementation rules for intelligent agents became enforceable in mid-July, creating a dedicated regulatory category with tiered authorization and filing requirements. Illinois enacted a state law requiring annual independent safety audits for large frontier developers. Delaware's contribution is different in kind: instead of regulating agent behavior directly, it proposes a legal vessel that assigns rights, duties, and accountability to the agent's activity. Together these moves show regulators approaching the same problem from several angles at once.
For enterprises, the practical value is optionality with guardrails. An AIC could let a company spin up an agent-run entity to transact in a bounded way, with logging, disclosure, and a member liability shield that has clear exceptions. That is a more governable path than letting agents act through existing corporate structures with undefined accountability. The sandbox's disclosure requirements also protect counterparties, which reduces the reputational and legal risk of dealing with an agent-operated business. The design tries to make autonomous commerce legible to the humans and institutions on the other side of a transaction.
The caveats leaders should hold onto
This is a proposal, not a statute. The legislation is expected to reach the Delaware General Assembly next year, and the details that matter most, capitalization thresholds, audit expectations, and the precise scope of the liability shield, could change materially in drafting. The 30-month sunset and the banking carve-out signal caution, and the admission committee's discretion means participation would be selective rather than automatic. Technology leaders should read the AIC as an early signal of where entity law is heading, not as a mechanism available for use this year.
There are hard questions the framework will have to answer. Who is accountable when an autonomous agent inside an AIC causes harm that falls outside the enumerated exceptions to the liability shield? How do activity logs get audited, and by whom, at machine speed? How does an internal-affairs doctrine built for human-run companies apply to one run by software? None of these are reasons to dismiss the effort, and all of them are reasons to engage early if agent-operated transactions are on your roadmap. The organizations that help shape these rules will understand them best when they take effect.
What to do with this now
The immediate action is to inventory where your agents already transact or could soon. Any agent that negotiates terms, commits spend, or accepts obligations is creating legal exposure today, well before an AIC-style entity exists to contain it. Map those flows, confirm which human or corporate entity currently bears the liability, and make sure your contracts and controls reflect that reality. The Delaware proposal is a prompt to get that accounting straight rather than a solution you can adopt.
It is also worth assigning someone in legal and engineering to track this file jointly. Agent accountability is becoming a cross-functional discipline, and the jurisdictions moving first, Delaware, China, and individual U.S. states, are setting patterns that will shape compliance for years. A company that understands the emerging entity and audit expectations can design its agent architecture to fit them, with logging, disclosure, and clear human ownership built in from the start. That preparation converts a regulatory unknown into a manageable design constraint.
