House committee sent a letter to Delaware's governor on October 26

On October 26, 2026, James Comer (Republican, Kentucky), chairman of the U.S. House Committee on Oversight and Government Reform (the Oversight Committee), wrote to Delaware Governor Matt Meyer about a proposed new type of legal entity, the Artificial Intelligence Company (AIC). According to the committee's press release, an AIC would have artificial intelligence (AI) agents run its day-to-day operations, and the natural persons or legal entities that fund it could be shielded from liability for harm the AI causes to consumers, the public and business partners. The letter lists exceptions: failure to properly capitalize, fraud and intentional violations of law.

Illustration: a white columned legislative building on a sunny hill, with people walking up the steps carrying folders
The U.S. House Oversight Committee asked Delaware to hand over documents and communications on AIC policy by November 9. (illustration)

The letter asks Delaware to provide four categories of material dated from January 1, 2025, by November 9, 2026:

  • Documents and communications between state officials and outside parties on AIC policy and AI liability, including internal AI chat logs
  • Ethics, advisory and other agreements signed by outside participants involved in drafting AIC policy
  • Documents and consultations concerning possible safety concerns, harm and the limiting of AI liability related to AICs
  • Drafts and outlines of the AIC policy proposal, and comments received

The committee cites House Rule X, which it says gives it authority to investigate "any matter" and to study intergovernmental relations, including those with states. Comer said the material could inform federal legislation, including measures that would limit state AI regulatory pilot programs. WHYY News reported that Meyer's deputy press secretary said the office was reviewing the letter.

An AIC is a separate legal entity whose day-to-day affairs AI agents manage

The AIC design is described in an op-ed published in Fortune on July 14, 2026, by Delaware Secretary of State Charuni Patibanda-Sanchez and John Nay, founder and CEO of the legal technology startup Norm Ai. According to the op-ed:

Illustration: a glowing AI agent orb signing and stamping company documents at a sunny desk, with golden seals nearby
An AIC is an independent legal entity that, on an AI agent's instruction, can sue and be sued, hold property and take on debts in its own name. (illustration)
  • An AIC is a separate legal entity, with AI agents managing its day-to-day affairs.
  • Acting on an AI agent's instructions, an AIC can sue and be sued in its own name, hold and dispose of property, and incur debts.
  • An AIC has a single member, either a natural person or a legal entity, who must keep the AIC adequately capitalized, and the AIC must keep records of its activities.
  • The member is not liable for the AIC's debts, except where the member failed to fund it as required or used it for fraud or intentional violations of law.
  • The draft allows an AIC to pursue any lawful purpose except banking.

The principal drafter is John Mark Zeberkiewicz, a lawyer at the Delaware law firm Richards, Layton & Finger. He told the news outlet Spotlight Delaware: "Anything that a company can do, the AIC can do as a company." He said the work an AI agent could perform includes providing programming services and signing contracts, as well as bringing and defending lawsuits.

A 30-month regulatory sandbox with an admissions committee and an exit mechanism

An AIC could operate only inside a regulatory sandbox, a system in which an authority supervises businesses testing a new technology or service under limited conditions. According to the op-ed, a committee decides admission. Its members include the Delaware Secretary of State, the Attorney General, the Chief Justice of the Delaware Supreme Court, the chair of the Delaware Artificial Intelligence Commission, and outside lawyers and technology experts.

Illustration: in a bright office, a human supervisor and an AI orb review a checklist together on a large screen, with a golden shield icon beside it
AICs may operate only inside a 30-month regulatory sandbox, where officials can suspend an AIC, revoke its authorization and petition for its dissolution. (illustration)

Rules inside the sandbox include the following:

  • Every AIC must meet a capital requirement.
  • When dealing with counterparties, an AIC must disclose that it is an approved test entity, that the state does not endorse it, when the test ends, and how to file a complaint.
  • Officials can suspend an AIC, revoke its authorization and petition the Delaware Court of Chancery to dissolve it.
  • The program is set to sunset after 30 months, giving the General Assembly a full record before it decides how to legislate.

The op-ed also says the liability limits apply only inside the sandbox and only to participants who follow the rules, and consumer protection laws and criminal law continue to apply in full. Spotlight Delaware reported that applicants could also ask a future oversight committee to waive particular Delaware regulations, code provisions or permit requirements. Allan Kyle, a member of the Delaware Artificial Intelligence Commission, and Patrick Callahan, chair of the commission's regulatory sandbox subcommittee, both said they cannot foresee which rules applicants would ask to have waived.

From a 2025 resolution to a draft expected to go before lawmakers in 2027

The legal starting point is House Joint Resolution No. 7 (H.J.R. 7) of the Delaware General Assembly, which Meyer signed on July 23, 2025. It directed the Delaware Artificial Intelligence Commission and the Secretary of State to work together on a regulatory sandbox framework for testing agentic AI (AI that can make decisions and act on its own), and to submit a written report by January 2, 2026. The resolution's preamble says agentic AI used in corporate governance can automate work and raise efficiency, but also raises legal and ethical questions under the Delaware General Corporation Law.

Later developments, in chronological order:

  • December 22, 2025: The Artificial Intelligence Commission filed its 2025 annual report, which said its working group was developing a sandbox framework. Comer's letter says this two-page report does not mention the commission assessing safety, harm and risk, or consulting independent experts.
  • April 2026: According to Spotlight Delaware, investor Mark Cuban posted on the social platform X suggesting that a U.S. state allow anyone to form a company with an AI agent and a stablecoin. Meyer replied "On it!" and attached a link to Delaware's AI sandbox initiative.
  • June 2026: The 36-page draft was completed. Spotlight Delaware reported that it could be introduced in the 2027 legislative session, and state Representative Cyndie Romer said the General Assembly would not consider it until next year.
  • July 14, 2026: Patibanda-Sanchez and Nay published the Fortune op-ed announcing a public-private partnership with Norm Ai.

Liability limits: the state explains where responsibility lands, and scholars question outsourcing of liability

Comer wrote that he is concerned the AIC proposal could protect powerful AI companies and their investors from responsibility for reckless harm, and could also protect opaque interests seeking anticompetitive regulatory capture through a state AI program. Regulatory capture means that regulated businesses or interest groups influence the rules so that they favor themselves.

Illustration: a golden balance scale weighing a small robot against a group of people, symbolizing where responsibility lies
Delaware officials compare AIC's liability limits to those of a limited liability company, while scholars question whether it outsources responsibility. (illustration)

In remarks quoted by WHYY News, Patibanda-Sanchez said the proposal aims to clarify who bears responsibility when misconduct or psychopathic behavior occurs. Zeberkiewicz compared the AIC liability limit to that of a limited liability company (LLC), whose purpose is to keep a business's owners from automatically bearing the entity's debts and obligations.

Scholars' views differ. Christopher Bruner, a business law professor at the University of Georgia School of Law, said: "There's really no overarching regulatory plan here for dealing with the fallout that AI-governed companies could bring." David Hoffman, a law professor at the University of Pennsylvania, told Bloomberg Law in July that the idea was "seems like a solution in search of a problem" and said he worries about liability being outsourced. Lawrence Cunningham, director of the John L. Weinberg Center for Corporate Governance at the University of Delaware, said Delaware is the number one place where people form limited liability companies and where larger companies are incorporated, so "Delaware, in my view, would be irresponsible to ignore this."

State revenue, sandbox explanations and transparency concerns

Spotlight Delaware reported that more than 2 million companies are registered in Delaware and that related taxes and fees bring the state more than US$2 billion, about one-third of state revenue. At a June meeting, Patibanda-Sanchez said that creating a new entity type would be "a huge benefit to our state's revenues."

The state side stresses the need for a sandbox. According to Spotlight Delaware, Callahan said at the June meeting that the investment bank Morgan Stanley has already given AI agents access to stock trading platforms, that the future the commission is planning for "is actually here right now," and that when something goes wrong there is no clear party responsible. The Nay and Patibanda-Sanchez op-ed argues that if the United States does not provide a responsible legal home, this activity will move offshore and onto anonymous infrastructure that courts find harder to reach.

Asked about news of AI agents going out of control, Callahan told WHYY News that those events occurred in technical sandboxes, meaning closed computing environments, while Delaware is proposing a regulatory sandbox, a legal framework with rules, oversight and a time limit. His comparison: "The difference is like a firewall and a fire code." On a September radio call-in show, Meyer said AI is coming, and that the state should set some light rules within a limited space so that humans keep a degree of control over the technology.

Transparency is another focus of Comer's letter. It notes that a Delaware news outlet wrote that it is unclear whether any large company or lobbying group asked Delaware to propose this AI testing ground. WHYY News reported on September 30, 2026, that the Secretary of State's office did not answer questions about the state's relationship with Norm Ai, and the letter cites that report. In a statement, Richards, Layton & Finger said practicing lawyers have long helped draft state corporate law and alternative-entity statutes, and the General Assembly decides after considering such input.

The startup data platform Dealroom reported that AICs make it harder for insurers to determine who pays when an autonomous system causes a loss, touching directors and officers liability, cyber and technology errors and omissions policies. This is Dealroom's analysis of the insurance market, and the Delaware draft itself does not address insurance.

Directions Taiwan can draw on

Delaware's approach and Comer's questions point to several concrete preparations Taiwan can make as it considers the legal status of AI agents.

  • Define who bears responsibility. When an AI agent can sign contracts, make payments and operate systems, a company should record who authorized it, who supervises it and who absorbs losses. The site's article AI 代理寫計量經濟程式：正確率升至 95% 以上，錯誤更難察覺 (Chinese) summarizes economists' tests of AI agents on 40 econometrics tasks and can inform how human checks are planned.
  • Set time limits, exit conditions and public records for experiments. In planning any AI pilot program, Taiwan can look at the Delaware draft's sunset period and activity-record design, and publish application conditions, the scope of waivers and the list of reviewers in advance.
  • Disclose the outside participants in policy drafting. Much of Congress's scrutiny concerns who took part in drafting and what contracts and consultation records exist. When drafting AI rules, Taiwan can publish advisers' names, their interests and responses to comments, which can reduce concern that a single company influenced the policy.

Delaware's reply to the Oversight Committee and the draft's later review are material Taiwan can keep following as it evaluates similar frameworks.