Category: Regulation | Reading time: 3 min
Delaware is weighing a new kind of business entity, an "AI Company," in which an AI agent runs day-to-day operations. New details reported by Ars Technica on October 27 show how far the draft goes, and Congress is already asking questions.
What the draft would allow
Per Ars Technica, the draft bill would let an AI Company pursue any lawful purpose except banking. It would sit inside a 30-month regulatory sandbox. The single human "member" behind the company would not be liable for the AI Company's acts, except in cases of undercapitalization, fraud, or willful violation.
Berkeley's Andrew Baker, quoted by the Financial Times and cited by Ars, called it "an overly generous liability shield."
Congress steps in
On October 26, House Oversight Chair James Comer wrote to Delaware Governor Matt Meyer requesting documents on the initiative, citing potential for harm and a lack of transparency. Comer has also hinted at federal limits on state AI rules.
Why it matters
If an agent can run a company and the human behind it is largely shielded, the agent's records become the main way anyone can reconstruct what happened when something goes wrong. Counterparties, regulators and courts will need to know which actions an agent took, under whose authority, and whether those records hold up. Whatever happens in Delaware, the question of who answers for an agent's actions is moving from theory into draft legislation.
Source: Ars Technica, House Oversight Committee