What: Delaware is drafting an Artificial Intelligence Company Act that would let an AI agent manage a single-member business entity on its own, a framework Boston College law professor Brian JM Quinn criticises for rewarding owners who supervise their agents less.

Why it is important: Existing statutes already recognise electronic agents as legal agents, so the bill's main effect would be to move liability away from the companies deploying AI agents rather than to bring those agents within the law.

Proponents of the draft Artificial Intelligence Company (AIC) Act argue that AI agents sit poorly within the legal system and that wrapping them in a corporate shell would make them legible to it. An AIC would be a single-member entity similar to an LLC, able to sue and be sued, hold property, sign contracts and incur debt in its own name, with an AI agent as its sole manager.

Quinn contends that the structure instead builds an almost impregnable liability shield for frontier AI firms such as Anthropic, OpenAI, Meta and xAI. Because extensive day-to-day oversight of an AIC would open avenues to pierce the corporate veil after an adverse event, rational owners would step back from supervision.

He notes that the Electronic Transactions Act has recognised electronic agents for more than two decades, underpinning ordinary internet commerce, and that principals should remain liable when their agents go rogue.

Should Delaware proceed, he recommends selective liability, a mechanism already used in smaller private corporations: the AIC's member would remain liable for any criminal or consumer protection violations committed by the entity or its agents.

IADS Notes: The accountability gap the author describes has already been identified on the operational side, where AI agents executing refunds, loyalty adjustments and purchase orders hold credentials and act on behalf of the business, yet often lack a named owner, a documented scope and auditable action logs (RH-ISAC, August 2026). Reported AI incidents rose from 106 in 2022 to 445 in 2025, most of them involving an agent taking an unanticipated path with nobody watching, and companies pairing transparency with executive accountability report a 15% median AI value impact against 7% to 10% for the rest (BCG, October 2026). The liability question also arises in payments, as agents paying on consumers' behalf through wallets such as Stripe Link, Shop Pay and PayPal will require clear rules on who answers when they err (an a16z analysis of agent wallets, October 2026). Walmart has built that control into its own systems: shoppers authorise each agent, set spending limits or require approval per transaction and can revoke access at any time, while agents are barred from payment card data (Reuters, October 2026). Amazon and other large etailers block agents altogether, citing returns and fraud exposure, with dispute abuse the most common fraud concern (a16z, October 2026).

Delaware drafts a bill for companies run by AI agents