Delaware Weighs Letting AI Agents Set Up and Run Businesses
Delaware is considering a bill that would let AI agents run most lawful businesses, prompting a Republican congressman to demand safety assessments.
According to TechRadar, the state of Delaware is considering a bill that would allow companies to be run by AI agents, adding to a business code that already draws companies with its flexible rules. The outlet reports that the proposal would let any lawful business be operated by an artificial intelligence company, with banking excluded.
TechRadar also reports that the bill sets up a 30-month regulatory sandbox, a period for testing the approach before any permanent rules are set. The report describes the sandbox as a testing window but does not lay out in detail what oversight would apply during it.
Rep. James Comer, a Republican, has written to Delaware's governor to raise concerns about the proposal, according to TechRadar. Comer is calling for safety assessments and has warned that federal legislation could follow if the results are not satisfactory. The report does not say what form that legislation might take or when it might be considered.
It remains unclear from the reporting how far the bill has progressed or whether the governor has responded. The episode puts a state-level experiment in AI governance at the center of a debate that is also playing out in Washington.
Background
Delaware occupies an unusual place in American business. Many large companies are incorporated there, largely because its corporate code is long established, its courts have deep experience with business disputes, and its legislature has repeatedly updated company law to suit firms. That reputation is one reason the state often moves first on corporate questions, and why a proposal like this draws attention beyond its borders.
Corporate law has long assumed that a business is run by people. Companies are legal persons that can own property, sign contracts and be sued, but they act through directors, officers and agents who are individuals. Those individuals owe duties to shareholders and can be held responsible when something goes wrong. Questions about who signs, who is liable and who answers to regulators sit at the center of any move to hand operational control to software.
AI agents are programs that pursue goals across multiple steps, such as researching options, negotiating, making purchases or managing accounts, often with limited supervision. Their capabilities have grown quickly in recent years, and many organizations have begun using them for routine tasks. Whether they can be trusted with the authority that management holds in a company is a separate and harder question.
Regulatory sandboxes are a tool governments use to let new products operate under relaxed or supervised conditions for a fixed period, so regulators can observe outcomes before writing lasting rules. Financial regulators in several countries, including Britain, have used them for financial technology. A time-limited window follows that pattern, though the scope and safeguards of any sandbox depend heavily on its written terms.
Banking is usually treated differently because deposits, payments and lending affect the wider economy and carry strict consumer protection and anti-money-laundering rules. The exclusion described in the reporting reflects that distinction.
Why it matters
The proposal tests a basic idea in corporate governance: that accountability should attach to identifiable people. If an AI agent makes a costly decision, a contract error or a consumer harm, the question of who is responsible becomes urgent. A state law could try to answer that question in advance, but courts, regulators and customers would still have to test whether the answer holds up in practice.
The story also highlights a tension in the American system. States have long competed to attract businesses with favorable rules, and they often act as laboratories for new policy. At the same time, federal lawmakers can legislate on matters that cross state lines, and technology companies operate nationally. A state experiment in AI-run companies could invite federal action, or it could become a model that other states copy or reject.
For businesses and consumers, the stakes are practical. Customers dealing with a company want to know whom to contact, how disputes are resolved and whether a human is ultimately accountable. Investors want clarity on governance and risk. Even if the bill never takes effect in its current form, the debate forces these questions into the open before the technology is widely deployed.
What to watch
The first signal will be the governor's response to Comer's letter and whether the state commits to the safety assessments he has requested. The content of those assessments, and who conducts them, will shape how much confidence observers can place in any eventual outcome.
Next, watch the bill's details. Key provisions would include how liability is assigned, whether a human must be named for legal and regulatory purposes, what reporting the sandbox requires, and how regulators can intervene if problems emerge.
Finally, look to Washington. Any federal bill would signal whether lawmakers see state-level experimentation as a model or a risk, and other states will be watching to see whether they follow Delaware's lead.
Viewpoints
Commentary from our columns: three angles on this story.
The Rulebook privacy, safety and regulation
The most important question in this proposal is not whether software can perform business tasks. It is who answers when those tasks go wrong. Law has always attached duties to people: directors who approve decisions, officers who sign filings, employees who can be deposed. An AI agent running a company would need an accountable human behind it, or the law would need to invent new forms of responsibility from scratch.
A sandbox can be a sensible way to learn, but only if the rules that apply during the testing period are clear and enforceable. Consumers dealing with an AI-operated firm should know how to complain, how to get a refund and which regulator will take the case. Those protections should exist on day one, not after the first serious incident.
The exclusion of banking is a telling detail. It suggests lawmakers recognize that some activities carry risks too large to test loosely. Safety assessments, as Rep. James Comer has requested, are a reasonable demand. The harder work is defining what a passing grade looks like before the experiment begins.
The Ledger the business angle
From a business standpoint, the appeal of an AI-run company is obvious: lower overhead, faster decisions and the ability to test ideas cheaply. A state that makes such structures easy to form could attract founders eager to be first. That is the kind of competition Delaware has long been good at.
But investors price risk, and a company without clear human leadership is harder to value and insure. Lenders, partners and insurers will want to know who can bind the company and who pays when something fails. Without answers, many of the cost savings could be eaten up by higher capital costs or by refusals to do business at all.
The sandbox structure gives entrepreneurs a runway, but it also gives skeptical counterparties a reason to wait. The businesses that succeed under this model may be the ones that build accountability into their structure early, rather than the ones that simply promise the most automation.
The Skeptic open questions and weak points
It is worth asking what problem this bill actually solves. The reporting describes a regime that lets an AI company run nearly any lawful business, yet the open questions are basic: who signs, who is liable, and who decides when the software is wrong. A proposal that leaves those answers vague is not yet a finished idea.
Calling the period a regulatory sandbox sounds reassuring, but sandboxes are only as meaningful as their exit criteria. If the terms for judging success are unclear, a test can end with a quiet extension or a permanent rule that nobody seriously examined. The most useful question for supporters is what evidence would show the approach failed.
Rep. James Comer's call for safety assessments is a fair place to start. Before Delaware or any state commits to this path, it should be able to explain, in plain terms, how a company run by software would be held to account.