Delaware wants to let artificial intelligence run companies. No board, no officers, just an AI agent signing contracts, buying and selling property, bringing lawsuits in the company’s name, and so on.
A preliminary proposal — presented by Delaware’s secretary of state and the founder of a company focused on automating legal compliance through AI — calls this new entity an artificial intelligence company, or AIC, and contemplates testing it in a regulatory sandbox. As academics who have studied AI in corporate governance, we find this move intriguing.
But without a clearer picture of how Delaware’s proposed sandbox and associated AIC would address the risks and accountability gaps that removing humans from the loop would create, the proposal remains far from roadworthy. Until a more holistically conceived regulatory plan for AI-driven companies emerges, prudence dictates preserving a central role in corporate governance for human judgment and conscience — imperfect as they may be.
A sandbox is a sensible way to explore uncharted legal and market terrain. It lets regulators be choosy about who enters, learn from what happens, and pull the plug if things go wrong.
Delaware has good reason to move first. Other countries or states (see: Texas) might otherwise establish looser rules and pull potential business away from Delaware. Corporate franchise fees fund a large share of Delaware’s budget, and high-profile departures by companies such as Tesla Inc. and Dell Technologies have sharpened the urgency.
Delaware could try to achieve much of this by simply amending its existing corporate statute. For instance, it could drop the requirement for a director to be a “natural person” and allow AI to serve on boards.
But that might jeopardize the state’s core franchise. A distinct entity hedges this risk, allowing Delaware to experiment without disturbing the traditional corporate form favored by most US publicly traded companies.
So, the broad contours of this initiative are sensible enough. But Delaware’s preliminary proposal leaves some important questions unanswered:
- What should an AIC look like?
- What safeguards should such an experiment adopt?
- What will happen if Delaware’s judiciary loses the ability to influence corporate behavior through its finely tuned system of fiduciary duties?
According to the preliminary proposal, an AIC would be managed entirely by an AI agent, with a single member — which could be a human or another entity — whose sole responsibility would be to ensure adequate capitalization. That member would face no liability exposure for the AIC’s debts unless the member failed to adequately capitalize the AIC or used it to commit fraud or otherwise willfully violate the law.
This would completely change how we think about corporate accountability. Reputational concern, professional duty, ethical constraints, and fear of monetary and nonmonetary consequences all restrain human directors and officers. These forces exist because directors are people with careers and consciences. Corporate law has spent centuries building fiduciary constraints around this human core.
The AIC proposal doesn’t explain what would replace this if the core were gone. The contemplated “public-private partnership” implicitly asks us to trust that the private sector can embed ethics and regard for law into an AIC’s governing algorithm to a degree that renders human directors redundant.
It appears that regulatory capital would represent the main mechanism for managing risk incentives and the main fund available to compensate those harmed by an AIC’s activities. The proposal has nothing to say, however, about how those organizing such entities might be deterred from doing so in a way that recklessly exposes the public to substantial risks of harm.
It rightly observes that the AIC entity wrapper wouldn’t affect consumer protection law or criminal law, suggesting that it would fall to these and other legal fields (and, one imagines, other states) to clean up messes that AICs create. But the preliminary proposal offers no specifics on how the interaction with these non-corporate laws might work and whether they could respond effectively.
As with all regulatory sandboxes, meanwhile, the details are crucial — and the preliminary proposal doesn’t supply them. The proposed framework is described as broadly available (except for banking), but it doesn’t say whether Delaware officials would allow a fully autonomous company in a genuinely hazardous line of business.
What must an AIC disclose about its activities, when should it disclose, and to whom should it disclose? Such specifics would appear critical to counterparties, as well as to the state — the aim being to inform future legislation. But the proposal also includes no specifics on this.
How much capital is enough? This appears to be the core structure for mitigating risk and addressing harms, but US corporate law generally hasn’t relied on capital adequacy to protect creditors outside of banking.
Consequently, Delaware officials presumably lack experience setting these numbers, and the right figures could vary significantly by industry. Insurance could supplement the capital cushion, but the proposal doesn’t address this, leaving one to wonder what kind, how much, and who would underwrite it.
The authority that Delaware officials would retain to revoke an AIC’s sandbox privileges provides cold comfort and still leaves critical questions unanswered:
- How would the kill switch work?
- Who would hold this responsibility?
- What information would they have in exercising this authority?
The industry has long anticipated a move toward AI corporate governance, and the impetus behind Delaware’s proposal makes sense. A regulatory sandbox is a reasonable, contained way to gather data before writing permanent law.
Delaware deserves credit for floating a proposal rather than waiting and reacting. But sandboxes, and laws modeled after them, can be structured well or badly — and Delaware needs to ensure that the proposed sandbox and associated AIC are embedded within a holistically conceived regulatory plan before road-testing AI-driven companies.
This Insight appeared as a co-authored op-ed by Christopher Bruner, Sergio Alberto Gramitto Ricci, and Martin Petrin, first published in Bloomberg Law. Copyright 2026 Bloomberg Industry Group, Inc. (800-372-1033) www.bloombergindustry.com. Reproduced with permission.