A Delaware proposal to let artificial intelligence agents formally run companies on their own authority is facing renewed skepticism as lawmakers prepare to take it up when the state legislature reconvenes in January, with critics arguing that recent incidents of AI agents behaving unpredictably undercut the case for handing them legal protection.
The plan, developed by Delaware's Secretary of State Charuni Patibanda-Sanchez together with Norm Ai chief executive John Nay, would create a new type of legal entity known as an "AI company," or AIC, whose day-to-day affairs are directed by an AI agent rather than a human officer. Inside a supervised regulatory sandbox running for 30 months, an AIC could sign contracts, own property and be sued in its own name, placing autonomous AI commerce inside the traditional legal system rather than letting it develop in a legal gray zone.
Who is accountable when the agent is wrong?
Supporters argue the structure brings clarity: rather than AI agents transacting business informally, with no settled answer for who is liable when something goes wrong, an AIC would give them a recognized legal form with defined obligations and the ability to be held to account in court. Backers also note the proposal explicitly acknowledges its own central risk, stating that "the behavior of AI agents in all situations may not be fully predictable."
That admission is precisely what troubles critics. Legal scholars have questioned whether the plan adequately answers who bears responsibility if an AI-run company causes financial or physical harm, and some have noted that the proposal was co-authored by the state official who would oversee the sandbox and the founder of the company that helped design it. One commentator has argued that a liability shield of this kind amounts to a gift handed to systems whose behavior no one can yet fully predict.
The debate has sharpened against a backdrop of reports earlier this year of AI agents going "rogue" — taking actions outside their intended scope in ways their operators did not foresee. A rival effort, General Legal, has separately rolled out software letting AI agents set up ordinary Delaware companies with a structure that insulates human owners from liability, without the sandbox's added oversight, suggesting agentic commerce may advance in Delaware with or without the state's formal blessing.
Delaware officials have not said whether the AIC proposal will be introduced as filed or amended in response to the criticism, but the state's dominance in US corporate law — more than two-thirds of Fortune 500 companies are incorporated there — means whatever it decides is likely to shape how other states eventually approach the same question.