A distinctly strange piece of corporate law is having a very loud week. On October 26, 2026, Representative James Comer of Kentucky — chairman of the U.S. House Oversight Committee — sent a letter to Delaware Governor Matt Meyer demanding documents about a state proposal to create a new corporate entity whose day-to-day manager would not be a human being. It would be an AI agent. The entity is called an Artificial Intelligence Company, an AIC, and the fight over it is a preview of every accountability argument your business will have about autonomous systems for the next decade.
The short answer, first
What's actually happening: Delaware, home to a plurality of America's incorporated businesses, drafted legislation (dated June 2, 2026, per MLQ's detailed legal review) that would let a company be managed day-to-day by an AI agent, hold property, sign contracts, and sue or be sued in its own name — while generally shielding the human owner behind it from liability, if the entity is adequately capitalized. It is a draft, not law: no AIC bill appears in the Delaware General Assembly's database as of October 28, and public reporting suggests lawmakers could take it up in January 2027. The sudden escalation is Comer's investigation, which frames the draft as a liability shield for "reckless, harmful actions" by AI and raises the specter of federal legislation limiting state AI regulation.
How an AIC actually works
The architecture is genuinely novel, which is why it deserves more than a shrug:
- One member, one AI manager. An AIC has a single member (an individual or another Delaware entity) who designates an AI agent as the entity's manager. The agent runs the company's business and must document each action in a signed record.
- The entity is the person — not the AI. Contracts, debts, lawsuits and property attach to the AIC itself; the agent is just its designated manager. If the AI "dies," the member has 30 days to appoint a replacement and runs things personally in the interim.
- Nearly any purpose, but no banking. An AIC may have any lawful purpose except banking. It needs a registered agent and a human contact able to receive legal notices.
- Mandatory honesty with counterparties. Before doing business, an AIC must disclose that it's in a regulatory test, identify its member, explain the liability protections, and state that Delaware does not endorse the experiment.
How an AIC compares with a plain LLC
| Feature | Standard Delaware LLC | Proposed AIC |
|---|---|---|
| Day-to-day manager | Humans | A designated AI agent |
| Legal personality | The entity | The entity (AI acts as manager) |
| Owner liability shield | Yes | Yes, if adequately capitalized |
| Disclosure to counterparties | Generally not required | Mandatory — must flag the test, the member, the shield |
| Sector limits | Few | Banking excluded |
| Activity logging | Not required by statute | Signed record of every action |
Where liability lands (and doesn't)
The draft uses familiar limited-liability architecture and then extends it to agency law: an AIC's debts belong solely to the AIC, and if the company is adequately capitalized, the member is not liable merely "because the AIC was the member's instrumentality or because agency principles might otherwise apply." The minimum capital is left to Delaware's secretary of state to determine.
The shield has real exceptions. A member can still be liable for fraud, willful violations of law, conduct outside the AIC's stated purpose that the member knew or should have known about — and the member owes the AIC fiduciary duties of loyalty and care. But as a Delaware Journal of Corporate Law analysis pointed out, the draft requires logs and records without guaranteeing that a warning ever reaches a human with authority to stop a transaction. That's the accountability gap in one sentence: supervision without a handbrake.
The AIC must consent to Delaware jurisdiction, with the Court of Chancery as the exclusive forum for sandbox-related disputes. Orderly for counterparties — but a judgment against a thinly capitalized AIC is still a judgment against nothing.
The sandbox, the politics, and the money
Participation would run through a 30-month regulatory sandbox chaired by the secretary of state, with a committee including the attorney general and the chief justice of the Delaware Supreme Court. Applicants must describe public and counterparty risks, show a data-security plan, prove adequate capitalization, and file a wind-down plan; sandbox status brings no immunity from civil damages or Delaware criminal and consumer-protection law. The sandbox flows from House Joint Resolution 7, which Governor Meyer signed on July 23, 2025, directing the AI Commission and secretary of state to build the framework.
The politics around it are telling. The proposal was developed through a public-private partnership with Norm Ai, a company that automates legal and regulatory compliance; its founder John Nay co-authored a July 2026 Fortune op-ed with Delaware Secretary of State Charuni Patibanda-Sanchez arguing that identifiable entities, activity logs, capital requirements and a shutdown mechanism make agentic commerce more — not less — supervisable. The state's economic motive is explicit: new entities mean new tax and fee revenue in the state that already houses a huge share of U.S. corporate registrations. And legal scholars are split: University of Delaware's Lawrence Cunningham has argued the state would be irresponsible to ignore AI-run companies, while University of Georgia's Christopher Bruner says the proposal lacks an overarching accountability plan.
Why this matters even if you'll never charter one
You will probably never form an AIC. But every business deploying agents — booking, invoicing, support automation, procurement — is quietly ratifying the same design questions Delaware is answering in statute form, the same way platforms like ServiceNow now demand signed credentials for every agent an organization runs:
- Who is accountable when an autonomous action harms someone? Your answer can't be "the tool." Define it in writing, per workflow.
- Can a human actually stop a transaction? Not "was a log generated" — is there a reachable handbrake before money moves?
- Can your counterparty tell they're dealing with an agent? Disclosure is becoming a formal expectation, not just good manners.
- Who's watching scope drift? "Conduct outside the stated purpose" is precisely what autonomous systems do when left alone.
- What sits behind the promise? Capital, insurance, or nothing? Courts and customers will eventually ask.
Autonomous commerce is coming either way — the bill's sponsors and its critics agree on that much. The companies that will benefit are the ones treating agent governance as an operating discipline now, while it costs documentation effort, instead of later, when it costs a lawsuit. If your organization is scaling agents without answers to the five questions above, that's exactly the gap our automation and governance work closes — talk to us.
Read next today
- Google opens Connected Apps to everyone: the agentic-web land grab
- Pichai's "one box": Search absorbed into an agentic layer
- Ro's Robin: the AI health agent that wants to quarterback your care
From the archive: ServiceNow's AI Passport: a signed credential for every agent
