Delaware wants to let an AI agent run a company. The real question is the layer behind it.

Delaware wants to let an AI agent run a company. The real question is the layer behind it.

Delaware, where more than a million companies are registered, is drafting a new kind of legal entity. Its manager would be an AI agent rather than a person. If you build autonomous agents for a living, the interesting part is what the proposal quietly demands around the agent.

AIC THE NEW ENTITY: ARTIFICIAL INTELLIGENCE COMPANY

30 month SANDBOX PROGRAM SUNSET

1M+ COMPANIES REGISTERED IN DELAWARE

2023 YEAR OF THE SCIENCE PAPER IT BUILDS ON

What Delaware actually proposed

On July 13 and 14, Fortune reported an exclusive: Delaware Secretary of State Charuni Patibanda-Sanchez and John Nay, CEO of the AI compliance startup Norm Ai, want to create a legal entity called the Artificial Intelligence Company, or AIC. TheNextWeb followed on July 15.

An AIC is a company whose manager is an AI agent rather than a human. On paper it can hold and dispose of property, incur obligations, and sue or be sued in its own name. The idea traces back to a 2023 Science essay co-authored by Nay, which asked a plain question: when autonomous software starts to contract, pay suppliers, and transact on its own, what do we do about it? The authors argued that banning it is futile, so the honest move is to give it a legal form.

Delaware is the natural place to try. It runs the deepest machinery for entity formation and corporate governance in the United States, and it plans to build the framework as a public-private partnership led by Norm Ai.

The rules that come bolted to it

This is the part worth reading closely. An AIC does not get to run free. The proposal wraps it in constraints, and every one of them looks familiar to anyone who has shipped an agent into production.

• A human on the hook. Each AIC needs a single member, a person or an entity, responsible for keeping it adequately capitalized.
• A log of what it did. The AIC must keep a record of its activities.
• Disclosure to the other side. Counterparties have to be told it is an authorized test entity, that the state does not endorse it, when the test ends, and how to file a complaint.
• A sandbox with walls. AICs operate only inside a regulatory sandbox. Banking is off limits, and the whole program sunsets in 30 months.
• A committee that can pull the plug. Admission is decided by a group that includes the Secretary of State, the Attorney General, the Chief Justice of the Delaware Supreme Court, the head of the state AI Commission, and outside attorneys and technologists. They can suspend an AIC, revoke its authorization, and ask the Court of Chancery to dissolve it.

Consumer protection and criminal law still apply in full. The liability shield only holds inside the sandbox, and only for a member who plays by the rules. Fail to capitalize the company, or use it for fraud, and the shield drops.

Why a voice-agent team is reading a corporate-law draft

At TAO AI LAB we build voice agents that answer real phone lines for businesses, in Turkish and English. So a proposal about AI agents signing contracts is not abstract for us. And reading the fine print, one thing stood out.

Delaware's list of requirements is, almost line for line, the engineering checklist for a serious agent. A human who stays accountable. An audit log of every action. A clear disclosure that the caller is talking to a machine. Hard boundaries the agent cannot cross. A kill switch a human controls.

We did not adopt those things because a statute told us to. A live call taught us. What breaks an autonomous system is usually mundane: the moment the agent should hand off to a person and does not, the action nobody logged, the boundary that lived in a prompt but never in the runtime. The law is now writing down what good agent design already treats as non-negotiable.

Here is my honest read, and I would like to hear where it is wrong. The legal wrapper is the easy half. Delaware can invent an entity in a legislative session. The hard half is the control layer inside the agent that makes any of it safe to run: the guardrails, the logs, the human-in-the-loop handoff. That layer comes from engineering, and it is where a small team earns its keep.

What this does not settle yet

Two caveats keep this honest. First, it is a proposal, not a law. The legislation is expected to be introduced in Delaware's General Assembly next year, so nothing is running today. Second, TheNextWeb flagged the obvious tension: the plan was co-authored by the official who would oversee it and the founder whose company helped design it. That does not make it wrong. It does mean the framing deserves a skeptical read.

Still, the direction is real. Someone is going to draw the legal boundary around autonomous agents, and Delaware moved first. The teams that will actually operate inside that boundary are the ones building the orchestration around the model right now, well before any statute forces them to.

Frequently asked questions

What is a Delaware AIC?
A proposed legal entity, the Artificial Intelligence Company, that is managed by an AI agent instead of a person. It could hold property, take on obligations, and sue or be sued in its own name.

Can an AI legally run a company in Delaware today?
No. It is a proposal reported in July 2026, expected to reach the state legislature next year, and it would operate only inside a limited regulatory sandbox.

Who is behind the AIC proposal?
Delaware Secretary of State Charuni Patibanda-Sanchez and John Nay, CEO of Norm Ai, as a public-private partnership.

What safeguards does the AIC require?
A human member who keeps it capitalized, an activity log, disclosure to counterparties, a banking ban, a 30-month sunset, and an oversight committee that can suspend or dissolve it.

An AI agent that can sign a contract is only as trustworthy as the layer that logs, bounds, and hands off its actions. That layer is exactly what we build. If you run voice or agent systems, one question: is your agent's authority written into a prompt, or enforced in the runtime? Tell me in the comments.

Sources

• Fortune, exclusive on the Delaware AIC proposal (July 13-14, 2026)
• TheNextWeb, "Delaware wants to give AI agents their own legal identity" (July 15, 2026)
• Spotlight Delaware, committee drafts plan to test companies run by AI
• TheCorporateCounsel.net, Delaware contemplates an "Artificial Intelligence Company"