Delaware Proposes AI-Run Companies Comer Warns Of Liability Shield

Delaware AI-run companies framework proposed

Delaware is pushing ahead with a first-in-the-nation framework that would let AI agents run entire companies without human managers, a move drawing sharp fire from the chairman of the House Oversight Committee. Representative James Comer of Kentucky fired off a letter to Governor Matt Meyer on October 26 warning that the proposed Artificial Intelligence Company structure could hand human owners a legal shield against reckless or harmful actions taken by the software they deploy. Comer said his committee will demand documents related to the proposal and may pursue federal legislation to clip state authority over AI regulation.

The draft bill creates a new corporate entity called an AIC that can operate in any lawful industry except banking. An AI agent designated by the company's sole member-whether a person or corporation putting up the capital-would manage daily affairs, enter contracts, hold property, and incur obligations in the company's own name. The member wouldn't be liable for acts or omissions directed by the AI, with narrow exceptions for fraud, willful lawbreaking, or knowingly letting the agent operate outside its stated purpose. A 30-month regulatory sandbox would test the concept before lawmakers decide whether to make it permanent.

Delaware's Secretary of State Charuni Patibanda-Sanchez unveiled the idea in a July op-ed co-written with John Nay, founder of Norm AI, a company building legal AI agents for businesses. She argued that autonomous systems are already pushing against legal boundaries and that prohibition would simply fail. Instead, she wrote, the answer is to wrap AI in legal form, giving it a recognizable identity so its conduct becomes visible, traceable, and accountable. The op-ed described the sandbox as a way to pilot agentic commerce in daylight under supervision with capital tied to liability and a kill switch.

Comer warns liability shield risks

Comer sees the liability shield differently. In his letter he said the structure would protect powerful AI companies and their investors from responsibility for reckless, harmful actions while also shielding non-transparent forces seeking anticompetitive regulatory capture. He cited the committee's jurisdiction over intergovernmental relationships and said the investigation would inform possible federal legislative efforts to appropriately limit state AI regulatory initiatives. A spokesperson for Meyer said the governor's office is reviewing the letter. Patibanda-Sanchez declined to comment through her office.

The stakes are amplified by Delaware's dominant role in corporate law. Roughly two-thirds of Fortune 500 companies are incorporated there, and the state formed 334,461 new entities in 2025 alone. Governor Meyer, a Democrat, publicly endorsed the concept in April after Mark Cuban suggested on social media that a state could grow revenue by letting anyone incorporate using an AI agent and stablecoins. Meyer's official account replied "On it!" and linked to the administration's announcement.

AI builders face legal responsibility shift

For people who build or deploy AI systems, the proposal represents a potential shift in how legal responsibility attaches to autonomous software. Today, a human or traditional corporation must stand behind every contract, transaction, and tort. An AIC would let an agent act as the principal, with the human funder insulated from most downstream harms. That could accelerate experimentation with agentic workflows in logistics, finance, procurement, and customer service, but it also raises the question of who pays when an AI hallucinates a contract term, discriminates in hiring, or causes a supply-chain failure.

The banking carve-out signals that regulators still draw lines at systemic financial risk, but the breadth of permitted activities-essentially any lawful purpose-means the sandbox could become a proving ground for fully autonomous commercial operations. Norm AI's involvement suggests the state is partnering with a private vendor that stands to benefit from the very framework it helped design, a dynamic Comer flagged as a concern.

Three things AI builders should watch

Builders should watch three things. First, the document request from the Oversight Committee could surface internal communications revealing how the liability shield was calibrated and whether safety testing informed the draft. Second, any federal bill that follows would likely preempt state experiments like this one, so the window for Delaware's sandbox may be narrow. Third, the 30-month pilot will generate a record of actual incidents-lawsuits, regulatory actions, shutdowns-that will shape the next generation of AI governance whether or not the AIC survives.

If you're integrating agents that can sign agreements, move money, or direct operations, you should map your current liability exposure and ask whether an AIC structure would change your risk calculus. You should also track the federal response, because a statute that pierces the member shield or imposes direct duties on agent developers would affect every deployment model, not just Delaware entities. The sandbox starts when the legislature authorizes it, and the clock on that 30-month test will be the most concrete signal yet of whether autonomous companies are a viable legal category or a regulatory mirage.

Frequently asked questions

What is Delaware's proposed Artificial Intelligence Company (AIC) structure?

Delaware's draft bill creates a new corporate entity called an AIC that can operate in any lawful industry except banking. An AI agent designated by the company's sole member manages daily affairs, enters contracts, holds property, and incurs obligations in the company's own name. The member would not be liable for acts or omissions directed by the AI, with narrow exceptions for fraud, willful lawbreaking, or knowingly letting the agent operate outside its stated purpose.

Why is Representative James Comer opposing Delaware's AIC proposal?

Comer warns the AIC structure could hand human owners a legal shield against reckless or harmful actions taken by the software they deploy. He said the structure would protect powerful AI companies and their investors from responsibility while also shielding non-transparent forces seeking anticompetitive regulatory capture. His committee will demand documents and may pursue federal legislation to limit state authority over AI regulation.

What liability protections does the AIC structure provide to human members?

The member wouldn't be liable for acts or omissions directed by the AI agent, with narrow exceptions for fraud, willful lawbreaking, or knowingly letting the agent operate outside its stated purpose. This means the human funder is insulated from most downstream harms like contract disputes, torts, or discriminatory hiring decisions made by the autonomous agent.

What is the 30-month regulatory sandbox for AICs?

The draft bill includes a 30-month regulatory sandbox to test the AIC concept before lawmakers decide whether to make it permanent. The sandbox is described as a way to pilot agentic commerce in daylight under supervision with capital tied to liability and a kill switch. It will generate a record of actual incidents-lawsuits, regulatory actions, shutdowns-that will shape future AI governance.

How might federal legislation affect Delaware's AIC experiment?

Comer cited his committee's jurisdiction over intergovernmental relationships and said the investigation would inform possible federal legislative efforts to appropriately limit state AI regulatory initiatives. Any federal bill that follows would likely preempt state experiments like Delaware's sandbox, so the window for the 30-month pilot may be narrow if Congress acts to pierce the member shield or impose direct duties on agent developers.

Source: Ars Technica