The Agent Gets a Legal Identity: What Delaware's AIC Draft Really Changes
The Agent Gets a Legal Identity: What Delaware's AIC Draft Really Changes
Most arguments about governing AI agents stay inside the software: what the agent may read, which tools it can call, where a person has to approve. Delaware is asking a different question. If an agent does the work, can it also be the company?
A draft bill from a Delaware committee would create an Artificial Intelligence Company, or AIC, a corporate entity whose day-to-day affairs are managed by an AI agent rather than a person. It is not law. It has not been introduced. But it has already drawn a congressional letter, and it is the clearest attempt so far to answer, in legal language, what an agent is.
What the draft actually creates
Under the draft, an AIC is a separate legal entity with a single member. That member, a person or another company, capitalizes the entity and receives its profits. The AI agent is designated to manage the business and affairs, and to conduct the business, of the AIC. The entity can sue and be sued in its own name, hold and dispose of property, and incur obligations. It may have any lawful purpose, with one exception: it may not be a bank.
The bill would not create AICs directly. It proposes a 30-month regulatory sandbox, an Artificial Intelligence Regulatory Sandbox, where applicants could test the structure under supervision. A committee including the Secretary of State and the Attorney General would accept or deny applications, and applicants could ask to be exempted from specific Delaware regulations, codes, or licensing rules.
The legislation is likely to be introduced in the General Assembly next year, according to Spotlight Delaware, which reported on the draft in June 2026. State Representative Cyndie Romer, who sits on the sandbox subcommittee, told the outlet the proposal would not be considered until next year.
The shield is narrow on paper
Limited liability is the point of most corporate forms, and the AIC is no different. The draft says the member "is not liable for any act or omission that the AIC takes or fails to take, or that another person takes or fails to take at the direction of the AIC."
That sentence is doing a lot of work, so read the exceptions. The shield does not cover a member who fails to keep the company adequately capitalized, uses it to commit fraud or a willful violation of law, or caused or allowed the AIC to conduct business the member knew or should have known was outside the company's statement of purpose.
In other words, the law would not ask a court to find the person who pressed the button. It would ask whether the company was funded, whether the work stayed inside its stated purpose, and whether the failure was intentional. Those are three questions any business running agents can answer now, without waiting for a statute.
Why Delaware would do this
Delaware is not a neutral observer in corporate law. More than 2 million companies keep their legal home there, and taxes and fees on them contribute more than $2 billion to the state budget, about a third of all revenue, according to Spotlight Delaware. In 2025 the state formed 334,461 new entities, and about two thirds of the Fortune 500 are incorporated there, Ars Technica reports.
A new entity type is a new fee line. "Should a new entity form be created out of it, that's really a huge benefit to our state's revenues," Secretary of State Charuni Patibanda-Sanchez said at a June 2026 commission meeting, according to Spotlight Delaware.
The intellectual case came in a Fortune op-ed in July 2026 by Patibanda-Sanchez and John Nay, founder and CEO of Norm Ai. "The answer is not to try to prohibit the activity," they wrote. "The answer is to wrap AI in legal form. Give an autonomous system a recognizable legal identity and you make it legible to law."
The pushback arrived before the bill
On 26 October 2026, House Oversight Committee Chairman James Comer (R-Ky.) sent Governor Matt Meyer a letter requesting documents on the proposal. "We are concerned that Delaware's AIC proposal could be used to protect powerful AI companies and their investors from responsibility for reckless harmful actions," Comer wrote, adding that the information could inform federal legislation to limit state AI regulatory initiatives.
Legal academics are skeptical for a different reason. Andrew Baker, a law professor at UC Berkeley, told the Financial Times the proposal looks likely "to be an overly generous liability shield, which seems unwarranted without some underlying policy rationale that hasn't been developed or justified well." Martin Petrin of York University told Bloomberg that other states will probably follow, and that a federal framework could help address safety concerns and a regulatory race to the bottom. Both quotes were reported by Ars Technica on 27 October 2026.
The part worth copying: the log
Strip away the politics and one design choice stands out. Under the proposal, an AIC must keep a log of its activities. The op-ed puts the purpose plainly: "You make the agent's conduct visible, traceable, and accountable."
That is the same discipline businesses need before any law arrives. Patrick Callahan, who chaired the Delaware committee that studied the issue, said at a June meeting that AI agents are already transacting, and that when something goes wrong there is no clear party being held accountable. He said the investment bank Morgan Stanley is already granting AI agents access to stock trading platforms, according to Spotlight Delaware.
You do not need a sandbox to adopt the boring half of this framework. Five things make an agent run legible:
- One named owner. A person, not a committee, who can answer for the agent's work.
- A statement of purpose. Written down, specific, and narrow enough that an unusual action stands out.
- An activity log. What the agent did, with what tool, on whose instruction, kept where a reviewer can read it without asking the agent.
- Deliberate limits. Spend ceilings, vendor lists, and a defined set of actions that always need a human.
- A review cadence. Someone reads the log on a schedule, not only after an incident.
None of that is a liability shield. It is the evidence a shield would depend on.
What happens next
The bill is a draft. It has not been introduced, and if it is, it will be one proposal among many. Comer has signalled a federal interest in limiting state rules. Other states may compete with their own versions. The 30-month sandbox, if it ever opens, is short enough that its first cohort will be the test case.
For a business, the practical question is not whether Delaware passes this. It is whether your agents could survive the questions the draft asks: who funded this, what was it allowed to do, and what did it record. Companies that can answer those questions already run agents the way the proposal imagines. Companies that cannot will find that the legal form is the easy part.
If you are working out how to run agents with clear ownership and a review trail, contact our team and we will talk it through.
Sources
- No human at the helm: Delaware plan would let AI agents run entire businesses (Ars Technica, 27 October 2026)
- Delaware committee drafts plan to test companies run by AI (Spotlight Delaware, 30 June 2026)
- Comer Investigates Potential for Harm, Lack of Transparency in Delaware Initiative to Create New Artificial Intelligence Companies (House Committee on Oversight and Government Reform, 26 October 2026)
- Fortune op-ed on giving autonomous systems a legal identity (Fortune, 14 July 2026)
- Draft bill: Artificial Intelligence Regulatory Sandbox and Company (PDF)
