Delaware’s attempt to create a new kind of company run by artificial intelligence agents is now under congressional scrutiny, with lawmakers questioning whether the structure could let human owners profit from autonomous decisions while avoiding responsibility when those decisions cause harm. The proposal has drawn a formal investigation from House Oversight Committee Chairman James Comer, R-Ky.
The proposed Artificial Intelligence Company, or AIC, would place an AI agent in charge of a business’s routine operations rather than a conventional human manager. Delaware’s draft framework envisions the entity holding property, signing contracts, taking on debt, and suing or being sued in its own name. The idea was later described as an “agent-managed company,” but its central premise remains the same: software would direct the company’s affairs while the people or organizations behind it receive a degree of legal separation.
In a letter to Governor Matt Meyer dated October 26, Comer asked Delaware to provide records concerning the proposal, including risk assessments, draft legislation, communications, outside agreements, and internal AI chat logs. The committee requested material dating back to January 2025 and set November 9 as the response deadline.
The congressional concern centers on accountability. The House committee says the plan could shield investors and powerful technology companies from liability for damage caused by an AI system, while also creating an attractive vehicle for businesses seeking to operate with minimal human oversight. Comer’s letter additionally questions whether industry participants helped shape the initiative without sufficient public disclosure or independent safety review.
The proposal grew out of work by Delaware officials and the state’s Artificial Intelligence Commission. A December 2025 commission report referenced a regulatory sandbox intended to test “agentic” AI in corporate governance, while draft legislation released in June described an AIC as an entity that could pursue nearly any lawful purpose other than banking. The proposed sandbox would run for 30 months, giving participating organizations time to evaluate the structure before broader adoption.
That sandbox is the policy’s most important—and most contentious—feature. Supporters could argue that a controlled experiment gives Delaware a way to attract AI startups and establish rules before the technology becomes widespread. Critics counter that testing liability limits in the real economy could expose consumers, employees, creditors, and business partners to harms that are difficult to reverse, especially if no clearly accountable human decision-maker is required.
For now, Delaware has not created a new corporate class, and the congressional inquiry could affect whether the proposal advances, changes, or stalls. The documents requested from Meyer’s administration may clarify who drafted the framework, which companies advised the state, what safeguards were considered, and how officials intend to assign responsibility when an autonomous corporate agent makes a disastrous call.
