A Critique of Delaware’s AIC Draft Legislation

Against backdrop of high-profile news about AI agents going rogue and perhaps even being an existential threat to humanity, the State of Delaware is preparing to take a drastic step. Proposed draft legislation known as the Artificial Intelligence Company (AIC) Act would allow for the creation of a brand-new business entity managed entirely by an AI agent, operating without routine human supervision. 

Proponents claim that wrapping AI agents in a corporate shell will make AI ‘legible’ to the legal system. In reality, the AIC Act builds a near-impenetrable liability shield for Big AI (frontier AI model firms, like Anthropic, OpenAI, Meta, xAI, etc.). Rather than forcing Big AI to keep their autonomous agents under strict control, Delaware’s proposal creates a framework that actively rewards companies for engaging in less oversight of their AI agents.

Proponents argue that as AI agents inevitably slip beyond human control, trying to ban them is futile, and that AI agents in the current form are an ill fit for the legal system and, as a result, beyond the law. Proponents of the AIC law propose that by wrapping AI agents in a corporate shell with legal personality one can overcome this infirmity. Under the AIC Act framework, an AIC can sue and be sued, hold property, enter into contracts, and incur debt in its own name. Statutorily, an AIC would be a single-member entity, similar to a single-member Limited Liability Company (LLC), with one crucial difference: its manager will be an AI agent and not a human being. The AI agent will manage the AIC on its own, without any human involvement. 

While proponents argue that the AIC makes AI ‘legible’ to the legal system, granting legal personality to the AIC also creates a strong liability shield that benefits the owners of the AI agent who are responsible for deploying them into the stream of commerce. At precisely the time when Big AI is warning the public that its models and AI agents have the potential to go rogue and cause great harm, the legislature is creating a new business structure that has, it appears, a single purpose: the protect Big AI from liability. 

In truth, AI agents are already legible under current legal frameworks and it is not necessary to grant them personhood. While AI agents may not have the capacity to act as agent under the common law, there are already statutes on the books, dating back more than two decades that recognize electronic agents as legal agents. Without the Electronic Transactions Act, we would have been unable to engage in Internet commerce, where the most common transactions involve humans routinely entering into contracts with electronic agents (ie websites). Electronic agents, including AI agents, are already legible to our legal system, even in the absence of a new business entity to grant them legal personality. AI agents that go rogue should still impose a liability on their principals. Facing the potential of liability for the acts of their agents, Big AI will have incentives to engage in a higher degree of oversight and perhaps be more risk averse in rolling out new agents. 

Notwithstanding the fact that AI agents are already legible, the legislature may still believe it important to create a new business entity for AI agents to improve AI agent visibility in the legal system. In that event, the legislature should revisit the wisdom of the liability shield for the AIC. Protecting an AIC’s member from liability associated with its actions creates incentives for the AIC’s member to engage in minimal direct oversight and regulation. To the extent an AIC’s member engages in extensive oversight and regulation of the AIC on a day-to-day basis, that creates avenues to pierce the corporate veil in the event an adverse event. Rational AIC members will, as a result, step back from necessary oversight.   

Legislators should consider creating selective liability for AICs. Under corporate law, it is well understood that a corporation and its shareholders may agree, in its certificate of incorporation, to assign certain corporate liabilities or categories of corporate liabilities to the shareholders. The use of selective liability is not uncommon in smaller, private corporations. A similar approach to limited liability should be applied to AICs, making the AIC’s member liable for any violations of the criminal law or consumer protection laws committed by the AIC or its agents. Use of selective liability will create incentives for the AIC’s member to engage in an appropriate amount of oversight and regulation of the AIC.

The author’s article, Legible but Not Accountable: A Critique of the Delaware Artificial Intelligence Company Act, is available here.

Brian JM Quinn is a Professor of Law at Boston College Law School.