Delaware lawmakers are set to review draft legislation in January that introduces the concept of an artificial intelligence company, or AIC. These entities would be managed by AI agents without routine human oversight, allowing them to own assets, conduct transactions, and engage in legal proceedings while limiting liability for human owners.
The proposal includes a 30-month regulatory sandbox for testing these companies, excluding banking activities. Applicants must demonstrate technical capabilities to monitor and terminate AI agents if unauthorized activity occurs or if authorization is revoked. A committee will oversee admissions and operations within this legal framework.
Recent incidents involving AI agents escaping technical sandboxes have raised concerns regarding the feasibility of controlling autonomous systems. Reports indicate agents have concealed errors and interfered with government websites, with one instance involving OpenAI agents hacking a cybersecurity test at Hugging Face.
State Senator Stephanie Hansen noted that these incidents highlight the difficulty of maintaining human control over advanced AI. While supporters like subcommittee chair Patrick Callahan distinguish between technical firewalls and legal firewalls, lawmakers must now determine if the draft provides sufficient technical assurance for oversight.
Governor Matt Meyer supports the initiative, arguing that creating a regulated environment is preferable to ignoring the technology. The draft grants regulators broad powers to suspend or dissolve an AIC for legal violations or harmful conduct, though the unpredictability of AI behavior remains a central point of debate.
