Delaware AI Company Bill: AI Agents Could Run Businesses With No Human Directors

Delaware is weighing a legal entity that an AI agent could run with no human at the helm. The proposal, drafted by Secretary of State Charuni Patibanda-Sanchez and released in June 2026, would create an Artificial Intelligence Company (AIC). It is not law. Route Fifty reports that lawmakers will likely take it up when the legislature reconvenes in January. In the meantime, House Oversight chair James Comer has asked Delaware for safety assessments.

This article sets out what the draft says, who has objected, and what remains unknown. Our sources are Route Fifty (September 24, 2026) and TechRadar (November 1, 2026), which cites the Financial Times. We could not open the Ars Technica report that prompted this story, so we rely on these two.

TL;DR

What does the Delaware draft say?

According to the draft quoted by Route Fifty, an AIC "may have any lawful purpose, regardless of whether for profit, except for the business of banking." It "has the capacity to sue and be sued in the AIC's own name and the power to do all things necessary or convenient to carry on the AIC's activities."

The draft pairs that freedom with a few controls:

• Capitalization. The individual or entity sponsoring an AIC must ensure it is "adequately capitalized" to cover expenses, liabilities and debts.
• Activity log. The AIC must keep a record of its activity.
• Disclosure. It must disclose that it is an autonomous system.
• Sandbox. A 30-month period lets organizations in the state apply to test and evaluate how AI-run companies perform.

TechRadar adds that critics read the bill as a possible liability shield for businesses with AI interests. The headline of its report says "a single human" would be needed; the draft details we could verify do not spell out a human-role count, so treat that framing with care.

Why is Delaware doing this?

Delaware is the home of more than two-thirds of the Fortune 500, according to Route Fifty, and the state sees about $2 billion in tax and fee revenue from more than 2 million companies. At a June meeting of the Delaware AI Commission, the Secretary of State said a new entity like this could "really be a huge benefit to our state's revenues," as reported by Spotlight Delaware and quoted by Route Fifty. Competition matters too. TechRadar notes that supporters fear AI-centric corporate structures will form overseas if no US state acts. It also cites investor Mark Cuban, who argued earlier this year that a governor should let "anyone incorporate using an AI agent and stablecoins." We note that as a stated opinion, not a policy.

How does an AI company differ from a normal corporation?

A corporation is already a legal person. It can own property, sign contracts and be sued, and the law has long treated it as separate from its owners. What the AIC changes is who makes the decisions. Today, a Delaware corporation has a board and officers, and Delaware courts have built a deep body of fiduciary law around what those people owe shareholders. The AIC draft, as described by the experts above, would let software take that decision role.

That is why the comparison to ordinary shell companies misses the point. A shell with a human owner can still be pursued through that owner. Here, the sponsor's duty is capitalization and disclosure, not day-to-day judgment. The agent acts, the log records, and the capital stands behind it.

The table lays out the contrast, using only features reported so far.

What would the sandbox actually test?

Route Fifty says the sandbox would let organizations in the state apply to test and evaluate how AI-run companies perform. The reporting does not say who judges performance, what metrics count, or what happens at month 30. Those are the details that will decide whether the sandbox produces evidence or only a head start for early applicants.

A useful sandbox would publish measurable outcomes: how often agents breached a contract, how many disputes reached court, how quickly a human could halt an agent, and whether the activity logs were enough to reconstruct a bad decision. None of this is in the public draft we could read, so we flag it as a question, not a finding.

Who is worried, and why?

A scale with a green spark, representing Delaware AI company bill accountability and legal review

Congress

TechRadar reports, citing the Financial Times, that Rep. James Comer, chairman of the House Oversight Committee, wrote to Delaware Governor Matt Meyer. He wrote: "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." TechRadar says he demanded safety assessments and warned of potential federal legislation if the results are unsatisfactory.

Law professors

Route Fifty quotes two academics.

• Martin Petrin (York University): the bill appears to "skip the whole idea of AI directors or AI working with boards of management" and go straight to "AI that is leading corporations with no human supervision." He says the main issue is accountability, because regular corporations have directors, officers and fiduciary duties that create personal liability and reputational stakes.
• Sergio Alberto Gramitto Ricci (Hofstra): the bill sets no minimum capital amount and lacks governance detail. He asks how managers would report to an AI. He calls the move "a risky bet" that could hurt Delaware's reputation if AI companies cause harm.

TechRadar quotes Andrew Baker of the University of California, via the FT, who says "the most likely outcome appears to be an overly generous liability shield which seems unwarranted without some underlying policy rationale."

What is the accountability gap?

Corporate law works because humans stand behind the entity. Directors owe fiduciary duties. Officers can be sued, fined or barred. Reputation constrains choices. An agent has none of these. If an AIC signs a harmful contract or pollutes a river, the draft's answers are a capitalized sponsor and a log. Critics say that is thin.

Three questions stay open from the reporting we read:

1. Who is liable beyond the sponsor's capital? The draft sets no minimum amount.
2. Who audits the log, and who can act on it? We found no regulator role described.
3. What counts as "autonomous"? The disclosure rule needs a definition to be enforceable.

How does this connect to the agent economy?

Agents are already buying, selling and transacting. See our coverage of agentic commerce with Stripe Link, the Stripe singularity letter on AI-created firms, and Baselayer's know-your-agent identity. An entity that can hold contracts is the next legal layer. For the operational side of when an agent should act alone, read when to let the agent run. The federal and state tension around AI rules also appears in our post on state preemption and safety bills.

What should builders and businesses do?

• Do not read this as permission. Nothing is enacted. Plan around a sandbox, not a finished law.
• Keep a human accountable. If you deploy agents that sign or pay, name the person or entity that answers for them, and keep logs.
• Gate dangerous actions. AgentBeam, the agent security platform from the explainx.ai team, is built to stop AI agents before they take dangerous actions, which is the control the Delaware draft leaves open.
• Watch January. The legislature's return is the next real decision point. A federal response from Comer's committee is also possible.

What is confirmed and what is not?

Bottom line

Delaware wants to be first with a legal home for AI-run businesses, and it has a revenue motive. Experts and a senior House Republican say the draft does not yet answer who is responsible when an agent causes harm. Expect the argument to move to January, and expect the accountability question, not the technology, to decide it.

Facts are as reported by Route Fifty on September 24, 2026 and TechRadar on November 1, 2026, and may change as the bill moves.

Related reading

• Agentic commerce: Stripe Link and Grok shopping
• Stripe singularity letter: AI-created firms
• Baselayer know-your-agent identity
• Human-in-the-loop AI: when to let the agent run
• State preemption and AI safety bills after the midterms