Delaware wants AI agents to incorporate—and make limited liability the experiment

Delaware is preparing a 30-month pilot for companies operated day to day by autonomous AI agents. The proposed entities could sign contracts, make business decisions and file lawsuits. Banking would be prohibited, and supporters say adequate capital, activity logs and a shutdown mechanism would keep experimentation visible and supervised. Legislation is expected in the state’s January session.

The corporate form matters more than the novelty of an agent chief executive. Limited liability normally separates a company’s obligations from the personal assets of shareholders and managers. An agent-managed company adds a new gap: when the system causes harm, was the responsible actor the model developer, deployer, owner, company or nobody with meaningful assets? Delaware’s draft would protect individuals if the entity is sufficiently capitalized and its actions logged. That could create a recoverable pool for victims—or an unusually efficient liability shield.

House Oversight chair James Comer has requested records from Governor Matt Meyer by November 9, citing consumer harm and regulatory capture. The inquiry is political scrutiny, not proof of misconduct. Comer also questioned the role of a corporate lawyer who has represented Meta and Norm AI, the legal-technology firm helping develop the proposal. Delaware’s incorporation fees finance a substantial part of the state, giving it an obvious incentive to establish the category before rival jurisdictions or offshore centres do.

A sandbox can be useful if it tests the hard variables: minimum capitalization tied to plausible loss, mandatory human controllers, immutable logs, insurance, disclosure to counterparties and clear rules for shutting an agent down. Merely observing incorporated bots for thirty months would generate activity without accountability.

The proposal exposes the central legal problem of agentic commerce. Existing companies already deploy models that negotiate and transact; refusing a new entity does not stop that. But granting legal personhood before assigning responsibility risks making autonomy a corporate-law technology for separating profit from harm.