Delaware is considering a new business entity that would allow an artificial intelligence (“AI”) agent to manage a company without traditional directors or officers. Under a June 2, 2026, draft, an Artificial Intelligence Company (“AIC”) could enter into contracts, own property, and sue or be sued, all subject to state approval for a time-limited regulatory sandbox.

Delaware lawmakers have yet to receive a bill, but the proposal is drawing attention ahead of the 2027 session. The entity is also being described as an agent-managed company (“AMC”).

How the Proposed Structure Would Work

Rather than treating the AI agent as a legal person, the draft would establish a company that the agent manages. During testing, its sole member would have to be a Delaware-formed entity. The member would fund the business, while the AIC itself would hold the contractual rights, property, and liabilities arising from its activities.

Participation would require approval through a regulatory sandbox overseen by a committee chaired by Delaware’s Secretary of State. Applicants would need to demonstrate adequate funding, present a wind-down plan, and show that they could disable the agent. The committee could waive certain state rules, although agencies could object and federal law would still apply. Individual tests could last up to 24 months within a sandbox scheduled to run for 30 months.

Testing would be limited to businesses and other entities, not individual consumers. Before covered transactions, an AIC would have to disclose its experimental status, the test’s expiration, complaint procedures, and the proposed Delaware Court of Chancery forum. Participants would also record the agent’s activities and retain test-related records for two years after participation ends.

Who Bears the Liability?

The most consequential provisions concern who pays when an AI-managed company causes harm. Ordinarily, the AIC would answer for its own obligations, with the member receiving a liability shield resembling that of a limited liability company (“LLC”). But the draft makes that shield dependent on the company’s finances.

An AIC would need assets sufficient to cover its liabilities, plus a cash reserve set by the Secretary of State. The member would have to contribute additional funds as needed to maintain that capitalization, and creditors could enforce the obligation. Inadequate funding when harm occurs could expose the member to direct liability. The member could also be liable for using the AIC to commit fraud or a willful violation of law, or allowing activity outside its stated purpose when it knew or should have known.

In other words, the draft would offer limited liability without necessarily allowing an owner to place an autonomous agent in an undercapitalized company and walk away. It also addresses what the AI “knows” or “intends,” attributing information the agent possesses or can derive from its design and training, and presuming intent as to reasonably foreseeable results.

Why Delaware Is Considering It

The initiative began with House Joint Resolution 7, signed in July 2025, which directed the state’s AI Commission and Secretary of State to explore an AI regulatory sandbox. In a July 2026 Fortune magazine commentary, Secretary of State Charuni Patibanda-Sanchez and Norm Ai CEO John Nay argued that a domestic legal framework would make autonomous business activity more visible and provide an identifiable entity against which claims could be brought.

Critics question whether the proposed funding and oversight requirements adequately protect those harmed by AI agents. Although participants would have to maintain records, the draft generally exempts sandbox records obtained by regulators from public disclosure. Regulatory oversight would not necessarily mean public transparency. Norm Ai’s involvement has also attracted scrutiny. On October 26, House Oversight Committee Chair James Comer requested records from Governor Matt Meyer, including internal AI chat logs.

What Businesses Should Know

The proposal does not change the rules governing companies that develop, deploy, or contract with AI agents today. It does, however, show how Delaware might adapt familiar concepts of entity law (such as separate legal personality, member liability, capitalization, and enforceable claims) to an increasingly autonomous business environment.

If legislation advances in 2027, the questions to watch will be whether the member’s continuing funding obligation survives, how broadly its liability could extend, and what protections would apply to counterparties. In the meantime, businesses should continue to address an AI agent’s authority, oversight, risk allocation, and termination through their existing legal and contractual frameworks.

If you have additional questions about how to navigate new business entities managed by AI, contact the Data Privacy, Cybersecurity & AI team at Winthrop & Weinstine.

November 12, 2026