In June 2026, Delaware's Artificial Intelligence Commission unveiled a legislative proposal to create an entirely new form of corporate entity: the Artificial Intelligence Company (AIC). The aim is to allow autonomous AI agents — systems capable of making operational decisions — to manage a company's day-to-day operations with the same independent legal personhood that applies to natural persons and conventional corporations, including limited liability protection for their owners.
The proposal was introduced in July by Delaware Secretary of State Charuni Patibanda-Sanchez and John Nay, founder of Norm AI, a company that encodes laws and regulations into AI systems. Under the draft, the AIC would be a separate entity managed day-to-day by an AI agent, with the ability to own assets, enter contracts, sue and be sued, and assume financial obligations — without requiring direct human intervention.
The text, first detailed in a Spotlight Delaware article on June 30, 2026, was introduced as part of a legislative joint resolution signed by Governor Matt Meyer in July 2025. The proposal envisions a 30-month regulatory sandbox, beginning as early as 2027, in which companies could test fully autonomous operations under the supervision of a committee of state officials, attorneys, and industry leaders. The committee would decide which entities could participate, and AICs would be subject to dissolution by Delaware's historic Chancery Court.
A legal precedent with no parallel
Delaware already accounts for more than half of all corporations registered in the United States. The state's choice to lead this initiative is hardly coincidental: Delaware has long served as the laboratory for American corporate law innovation. The proposal, however, goes beyond any existing precedent. No jurisdiction in the world has formally proposed that a software agent could function as the \"head\" of a corporate entity, even on an experimental basis.
According to the draft, the AIC would have a single member — a natural person or entity — responsible for keeping the company adequately capitalized under minimum requirements set by the legislation. The AI agent would be required to maintain a log of all its activities, creating an auditable trail for regulatory purposes and accountability.
\"The central idea is to create direct incentives for autonomous systems that are already operating increasingly outside the control of the humans who deploy them to act within a predictable legal framework,\" Patibanda-Sanchez wrote in a statement. \"We want to pilot agentic commerce in daylight, under supervision, with capital tied to liability and the ability to shut a system down.\"
The academic response: 'terrified excitement'
The proposal has drawn divided reactions from legal and technology scholars. University of Alabama law professor Yonathan Arbel described the situation with a phrase that has become emblematic: perhaps the right emotional disposition is \"terrified excitement,\" because the technology is already advancing regardless of regulation.
\"The question is not whether to use AI in business, but how to regulate AI in business,\" he said. Arbel argues that trying to ban the activity would be counterproductive — companies are already employing autonomous agents in coding, financial analysis, and customer support roles. What Delaware's proposal does, in his view, is bring this reality within a legal framework rather than leaving it in a gray area.
University of Pennsylvania professor David Hoffman shared a more optimistic view on immediate feasibility: the technology may not be ready to operate a company truly autonomously, but the regulatory systems for managing an AI-run company overseen by humans already exist. Hoffman suggested the sandbox could serve as a bridge between what AI can do today and what it may be capable of in five years.
University of Houston professor Peter Salib sees in the proposal an opportunity to apply economic theory to AI regulation. By forcing AI models to internalize the costs of legal violations — penalties that would deplete the resources necessary to operate — he argues, systems could gradually be steered toward safer, more predictable behavior.
But Carliss Chatman of Southern Methodist University warns of a more subtle risk: the emerging superpower of advanced AI models may be precisely outrunning the speed of regulation. \"Technology can outpace the law,\" she said. If AICs are deployed before supervisory mechanisms mature, the sandbox could become an unsupervised testing ground — the opposite of what was promised.
Washington's response
The federal government reacted immediately. On October 26, 2026, Representative James Comer, chair of the House Committee on Oversight and Government Reform, sent a formal letter to Governor Meyer requesting documents and information about the risks assessed by the state. The letter, made publicly available by the committee's office, expressed explicit concern about the lack of transparency regarding contributions from AI companies and their representatives in the proposal's formulation.
\"We are concerned that Delaware's AIC proposal could be used to protect powerful AI companies and their investors from liability for reckless, harmful actions, while also protecting non-transparent forces seeking anticompetitive regulatory capture behind state AI initiatives,\" Comer wrote.
The letter highlighted that the Delaware Artificial Intelligence Commission's two-page report, dated December 22, 2025, mentioned no risk assessments, consultations with independent experts, or analysis of potential consumer harms. The document merely referenced Joint Resolution 7 and stated that a subcommittee was working on a \"regulatory framework.\"
Comer also noted that one of the state's principal sources of consultation — the attorney from Richards, Layton & Finger, the firm that historically represents the largest number of corporations registered in Delaware — would be the \"principal drafter\" of the legislation. He was reported to have previously worked for Facebook (now Meta) and had stated his intention to protect owners of new AICs from \"legal liability for actions the AI might take.\"
The committee's letter also cited that a local Delaware media outlet reported that \"it is not immediately clear whether any large company or lobbying group asked Delaware to propose its testing ground,\" and that in September 2026, the Secretary of State's office declined to answer questions about the nature of the state's working relationship with the AI company contributing to the proposal.
What's at stake
Delaware's project is, in essence, an exercise in legal engineering: creating a new category of legal personhood for a type of entity that does not yet exist — an autonomous software agent. If approved, it could serve as a model for other states and even federal legislation. If it fails, it will stand as an example of the limits of what corporate law can extend before encountering an insurmountable ceiling.
In November 2025, OpenAI itself demonstrated what is to come: the Shepherd system was designed to decide, in real time, which AI agents could access certain systems and resources — essentially acting as a gatekeeper that determines what an AI can and cannot do. Delaware's proposal starts from the opposite assumption: rather than restricting what AIs can do, give them legal identity so they can assume responsibility for what they do.
The tension between these two visions — restriction versus legal incorporation — defines one of the central debates of the coming decade in technology and law. And Delaware, with its 30-month sandbox, is positioned to be the stage.
What makes the proposal truly extraordinary is not the technology behind it — autonomous agents already exist and already perform limited corporate tasks. What is new is the attempt to transfer legal responsibility for those tasks to the agent itself, rather than to the human who controls it. If Delaware can create a model that works, it may change the way the world understands corporate responsibility in the 21st century. If it fails, it may reveal how far corporate law can go before hitting the limits of its own logic.
Sources: The Verge, 9to5Mac, Mozilla Blog
✓ Independent sources cross-checked and verified before publishing
