Delaware Floats Legal Framework for Companies Managed Entirely by AI

A Corporate Structure Built Around Software Autonomy

Delaware is preparing to introduce legislation that would permit artificial intelligence agents to manage entire businesses without human oversight - a proposal that reframes long-standing assumptions about corporate governance. The draft bill, outlined in July 2026 by Secretary of State Charuni Patibanda-Sanchez in a co-authored opinion piece, would establish a new entity type: the AI Company, or AIC. Under this structure, an autonomous system would direct daily operations, enter contracts, transact with suppliers, and handle financial obligations independently.

The proposal emerges from a recognition that existing corporate statutes in multiple US jurisdictions do not explicitly bar software from managing a company. Rather than attempt prohibition, Delaware's approach is to grant AI agents formal legal identity, creating what the secretary describes as a "defined target to which responsibility and damages can attach." The goal is visibility and traceability - attributes that depend on embedding autonomous systems within established legal frameworks rather than allowing them to operate in regulatory grey zones.

At Opentechwire, we have tracked the steady expansion of agentic AI across operational domains - customer support, logistics routing, procurement - but those deployments typically sit within human-supervised workflows. Delaware's draft moves the boundary: it contemplates software that does not escalate decisions to a person, and that carries legal personhood in its own right.

How the Proposed Structure Allocates Risk

The AIC would function as a distinct legal entity, able to sue and be sued under its own name, hold property, and incur obligations. A single "member" - either an individual or a corporate entity - would be required to capitalise the AIC and maintain a log of its activities. That member receives any profits but delegates management authority entirely to the AI agent.

Crucially, the draft statute shields the member from the AIC's debts and liabilities in most scenarios. The member is not responsible for acts or omissions undertaken by the AI, or by third parties acting at the AI's direction. Two exceptions carve out personal exposure: if the member uses the AIC to commit fraud or a wilful violation of law, or if the member permits the AIC to operate outside its stated purpose while knowing - or having reason to know - that the conduct falls beyond the scope.

This liability firewall mirrors the limited-liability logic of traditional corporations, but applies it to a context in which the decision-maker is software. The structure creates an incentive to capitalise the AIC adequately, because undercapitalisation pierces the shield. It also imposes a gatekeeping obligation: the member must define a clear statement of purpose and monitor whether the AI stays within bounds. How that monitoring occurs - given that the premise is minimal human involvement - remains an open design question.

Accountability Through Legal Identity

The rationale offered by Delaware's secretary of state centres on legibility. Granting an AI agent legal form makes its conduct "visible, traceable, and accountable," turning what might otherwise be opaque algorithmic behaviour into something that courts, regulators, and counterparties can address through existing legal mechanisms. If an AIC breaches a contract, the counterparty can sue the entity. If it fails to pay a supplier, the supplier has a named defendant. If it violates a regulation, the regulator has a target for enforcement.

This logic assumes that wrapping software in a corporate shell produces better outcomes than leaving autonomous agents unincorporated. The alternative - software acting on behalf of human principals in ways that blur attribution - could fragment accountability further, especially as agent capabilities scale. Delaware's proposal bets that a defined entity, even one run by an algorithm, is preferable to distributed, hard-to-trace decision-making.

Yet the framework also introduces a new class of counterparty risk. Businesses transacting with an AIC will need to assess whether the entity is adequately capitalised, whether its statement of purpose is narrow enough to constrain behaviour, and whether the member is monitoring effectively. Contract law has long dealt with corporate counterparties of varying creditworthiness; the novelty here is that the operational judgment sits with software whose reasoning may not be auditable in the same way a board resolution is.

Implications for Venture Capital and Corporate Governance

The proposal arrives at a moment when venture-backed startups are experimenting with agent-driven workflows - tools that book meetings, negotiate terms with vendors, or allocate cloud compute budgets. Most of these deployments retain a human approver for material decisions. Delaware's AIC structure would permit a founder or investor to fund an entity, define its mission, and step back, letting the AI handle execution without further sign-off.

For venture capital, the model offers a way to test autonomous business logic at lower overhead. An AIC could operate a micro-service - say, a programmatic arbitrage desk, a subscription management platform, or a logistics coordinator - without the salary burden of a management team. Returns would flow to the member; operational risk would sit with the AIC's capital buffer. If the experiment fails, the member's exposure is capped, assuming proper capitalisation and no fraudulent conduct.

The trade-off is transparency. Board meetings, management reports, and email trails create a record that investors and regulators can inspect. An AI agent's internal state - its reward function, its training data, its decision tree - may not be as accessible, even with an activity log. Delaware's draft mandates logging, but does not specify format, granularity, or auditability standards. Those details will determine whether the activity log serves as a genuine accountability mechanism or a compliance formality.

Regional Context and Regulatory Arbitrage

Delaware has long been the domicile of choice for US corporations, thanks to a mature body of case law and a specialised Chancery Court. The state's move to accommodate AI-run entities continues that tradition of legal innovation tailored to business demand. But the proposal also raises questions about regulatory arbitrage. If Delaware permits AICs while other jurisdictions do not, will autonomous entities cluster in Delaware, and will that concentration create enforcement challenges for regulators in other states or countries?

Across Asia, jurisdictions have taken varied approaches to AI governance. Singapore's Model AI Governance Framework emphasises human oversight and explainability; China's algorithmic recommendation regulations require disclosure of recommendation logic; South Korea's AI ethics guidelines stress transparency. None of these frameworks currently envision fully autonomous corporate entities. If Delaware's legislation passes, it may prompt other commercial hubs - Singapore, Hong Kong, Dubai - to consider whether they need parallel structures to remain competitive, or whether they prefer to maintain human-in-the-loop requirements.

The cross-border dimension matters especially for venture capital. A fund based in Singapore or Seoul might back a Delaware AIC to access US markets, but face complications if the AIC's autonomous conduct violates regulations in the fund's home jurisdiction. Export controls, data-protection rules, and anti-money-laundering obligations all assume a human decision-maker who can be held accountable. Adapting those regimes to accommodate AI agents will require coordination that extends well beyond Delaware's statute book.

Open Questions on Fraud, Intent, and Wilfulness

The draft statute preserves member liability for fraud and wilful violations, but applying those concepts to AI-driven conduct is not straightforward. Fraud typically requires intent to deceive; wilfulness implies knowledge and conscious disregard. If an AIC misrepresents a material fact in a contract negotiation because its training data contained biased examples, does that constitute fraud by the member? If the member did not review the agent's training corpus, does that amount to wilful blindness?

These questions will ultimately be resolved in litigation, but the answers will shape how much protection the liability shield actually provides. A narrow interpretation - holding members liable only when they directly instruct the AI to commit fraud - would make the AIC an attractive vehicle for risk-taking. A broad interpretation - imputing the AI's conduct to the member whenever the member could have foreseen the risk - would erode the shield and push responsibility back onto human actors.

Delaware's Chancery Court has a long history of developing corporate law through case-by-case adjudication. The AIC structure will likely follow the same path, with early disputes setting precedents on questions of intent, oversight, and adequate capitalisation. Investors and technologists watching the proposal should expect that the real rules will emerge not from the statute itself, but from the first wave of lawsuits.

What Comes Next

The draft legislation has not yet been introduced in Delaware's General Assembly, and the timeline for passage is unclear. If enacted, the AIC structure would be optional - one corporate form among many. Its adoption will depend on whether entrepreneurs, investors, and legal advisers see it as solving a real problem or creating new uncertainty.

The proposal reflects a broader tension in AI governance: whether to regulate the technology directly, or to regulate the entities and individuals who deploy it. Delaware's approach sides with the latter, treating legal form as the lever for accountability. Whether that proves sufficient will depend on how well activity logs, capitalisation requirements, and member oversight function in practice - and on whether courts are willing to pierce the corporate veil when autonomous conduct crosses legal lines.

For now, the draft remains a signal of intent. Delaware is betting that jurisdictions able to provide legal clarity for autonomous systems will attract the next generation of AI-enabled businesses. Other states, and other countries, will need to decide whether to follow, or whether human oversight remains a non-negotiable feature of corporate governance.