Delaware is testing whether an AI agent can be more than software. It wants to give one a company-sized legal shell, and the liability question is the hard part.
A Delaware committee studying artificial intelligence has drafted legislation for a new business entity called an Artificial Intelligence Company, or AIC. The idea is blunt: let AI agents do work normally handled by human executives, from signing contracts to filing lawsuits, inside a legal wrapper built for them. That is not a small tweak. Delaware is where more than two-thirds of Fortune 500 companies are incorporated, according to the state’s own corporate pitch and long-running corporate law data, so its experiments have a way of becoming everyone else’s paperwork.
The measure was drafted by corporate attorney John Mark Zeberkiewicz of Richards, Layton & Finger, one of the Delaware firms regularly involved in shaping the state’s corporate code. It grows out of Delaware’s AI Commission, which has been working with the Secretary of State’s office on a regulatory sandbox for agentic technology. Corporate governance was the first sector chosen for testing. Of course it was. Delaware makes a large share of its revenue from incorporation fees and franchise taxes, and it has no interest in watching Texas, Nevada or another rival state become the default home for companies built around autonomous agents.
Under the draft plan, an AIC would work something like an LLC. Owners would not be automatically liable for debts or obligations incurred by the entity, even if an AI agent, rather than a person, is running the operation. That is the bargain Delaware is trying to test. A business would apply to an oversight committee including the Secretary of State and the Attorney General, then operate inside a sandbox where officials can watch how an AI-run company behaves under real legal conditions. The bill is not expected to reach the General Assembly until next year.
Here’s the thing: a sandbox sounds tidy until you imagine the first contract dispute. If an AI agent signs a supply agreement, misses a deadline and triggers a lawsuit, the injured party will not care that the experiment was interesting. They will want someone solvent to pay.
The compliance business is already moving
Norm Ai, a New York startup that turns legal rules and regulatory frameworks into AI agents, has just closed a $120 million Series C at a $1.2 billion valuation. Khosla Ventures led the round, its first investment in the company, with returning backers including Blackstone, Bain Capital Ventures, Coatue, Vanguard and TIAA. The company says its customers collectively represent more than $30 trillion in assets under management, across banks, hedge funds and insurers. Blackstone is both an investor and a customer.
That funding round fits the Delaware proposal almost too neatly. Norm Ai runs an affiliated law firm, Norm Law LLP, where senior attorneys supervise AI agents instead of managing a pyramid of associates. It charges for outcomes, not billable hours. That is the model. If Delaware lets AI agents sign contracts and defend lawsuits for a company, someone will have to prove those agents are following the law while they do it.
Norm Ai wants to be that layer. Not the chatbot on top. The compliance machinery underneath. For founders, that distinction matters because an AI-native company will need more than a clever agent and a Delaware filing. It will need records, supervision, audit trails and a human professional somewhere close enough to be blamed when the software gets the law wrong.
The liability shield is the real fight
Ann Lipton, a business law professor, has raised the sharper question: if Delaware authorizes a nonhuman entity and gives its human investors protection, why would a court in another state honor that shield when one of its own residents brings a tort claim? That is where the AIC stops being a corporate law novelty and becomes a national legal problem. Delaware can write the entity. It cannot command every judge in America to treat it kindly.
Lipton is not reassured by AI’s public record either. She has pointed to systems making headlines as cheating assistants, plagiarism machines, fabulists and, in tragic cases, suicide coaches. That is not a comfortable resume for something about to be handed authority to sign binding contracts. You do not need to be anti-AI to see the problem. You only need to have read a complaint.
Frankly, the skepticism is fair. Delaware has rewritten its corporate code before to keep pace with new business models, and it often wins those bets. But a liability shield only works if courts outside Delaware respect it. Nothing in the draft legislation can guarantee that. For founders building AI-native startups, the AIC is worth watching closely, not adopting on faith.
The sandbox exists because nobody knows yet whether an AI-run company can survive contact with a real courtroom. Not Delaware. Not Norm Ai. Not the investors funding the compliance layer around it. The open question is not whether AI agents can execute tasks. They can. The question is who pays when execution becomes damage.
Also read: SBI and Solana Foundation Team Up to Build Japan's Onchain Financial Market • A Shanghai Chip Startup Is Betting 3D Stacking Can Break Nvidia's Grip • Amazon Bets On TwelveLabs To Turn Video Into The Next AI Battleground
