A customer emails a dealership to sell her vehicle. After some back and forth, the sales representative makes a $20,000 offer, which the customer accepts. They agree to meet at the dealership to finalize the deal. But when the customer arrives, she is told the deal is no good. Despite having an agreement over email for the sale, the customer is informed an artificial intelligence (AI) sales agent misquoted her, at which point a human sales representative tries to present a new offer. This is no hypothetical; it happened earlier this year. (As a post-script, hoping to avoid the resultant bad press, the dealership later agreed to honor the AI agent’s first offer.)
 
Traditional AI chatbots are limited in their operation: they cannot do more than respond with text when presented with text. They also lack the inherent common sense and business acumen a sales agent is expected to exercise. But AI agents are more sophisticated. And businesses are increasingly integrating AI agents into their workflow to perform customarily human functions.

All this begs the question: when AI acts for a business, who bears responsibility for its actions?

What Are AI Agents?

Unlike earlier AI chatbots that just generate text in response to prompts, modern AI agents are being used in various industries. These AI agents are tasked and trained to independently pursue objectives and to use tools to interface with existing company software. AI agents can work on tasks for hours, with little or no human intervention.

AI agents first became popular in January 2026, when open-source AI agent Clawdbot (now OpenClaw) went viral for its ability to help users perform basic tasks, including organizing computer files or placing online orders. More recently, Lyzr Inc., a startup that builds AI agents, used its own AI to draft dozens of investment memos and respond to queries from more than 100 investors—all while raising $100 million at a $500 million valuation. 

These AI agents have now begun to take on (and succeed) in performing more sophisticated tasks. And as AI agents are able to perform the same tasks as human agents, the next question becomes whether the law can appropriately account for the ramifications of unintended AI agent decisions.

Agency Matters: How a '90s Law Addresses Modern Day AI Problems

Despite a general belief that lawmakers are behind the curve on this type of emerging technology, legislators were addressing issues of agency as far back as the late 1990s.

In fact, the above hypothetical about whether an AI agent can bind its user to a contract was contemplated in the Uniform Electronic Transactions Act of 1999 (UETA). The UETA, adopted by 49 states, Washington D.C., and the U.S. Virgin Islands (with the exception of New York, although the New York City Bar Association published a report this year encouraging the state to adopt it), was originally published to harmonize state law regarding the validity of electronic transaction records and signatures.

UETA states that a contract may be formed by the interactions of the “electronic agents” of the parties—“even if no individual was aware of or reviewed the electronic agents’ actions or the resulting terms and agreements.” To avoid being bound by the electronic agent’s actions, an individual must: show that the electronic agent did not provide an opportunity for the prevention or correction of the error; and on learning of the error, the individual must promptly notify the other person of the error, state that the individual did not intend to be bound by the electronic record, and take reasonable steps to return to the other person the consideration received.
 
But how do AI agents—something unheard of in 1999—fit into the legislative picture?Incredibly, the authors of the UETA had the prescience to leave comments on exactly that topic: “While this Act proceeds on the paradigm that an electronic agent is capable of performing only within the technical strictures of its preset programming, it is conceivable that, within the useful life of this Act, electronic agents may be created with the ability to act autonomously, and not just automatically … If such developments occur, courts may construe the definition of electronic agent accordingly, in order to recognize such new capabilities.”

Did My AI Agent Do That?

Aside from entering into a binding agreement, AI agents can potentially ensnare their principals in civil liability. And no comedically timed “Did I do that … ?” by beloved Steve Urkel will save companies from what could be costly ramifications.

Recognizing the potential for AI agents to perform potentially tortious action, California has taken things a step further. In fact, the recently enacted California Civil Code 1714.46 provides strict liability for users of AI agents, stating that “it shall not be a defense, and the defendant may not assert, that the artificial intelligence autonomously caused the harm to the plaintiff.” Unlike human agents, whose actions may not be imputed on their principals under certain circumstances—i.e., where they exceed their scope of authority, or they “frolic” from their authorized tasks—California has preemptively rejected an AI agent’s autonomy as a shield for accountability.

Conclusion

For decades, the law treated software as a tool to be used by human operators. But modern AI agents are forcing courts and lawmakers to reconsider whether AI should instead be treated more like a human agent—able to bind its principal. And if so, is an entirely new legal framework needed? For now, most states may hold companies to the word of their AI agents in contracts, and at least California has decided to hold its citizens strictly liable for other harms that may arise. As both the technology and the law evolve, businesses deploying AI agents should assume courts will look first to the human or company that put the AI agent into the world, not to the agent itself. Simply claiming “my bot did that, not me” will likely not provide the protection most companies would hope to exist around the use of AI agents.

"Did I Do Bot?! When AI Agents Create Real-World Consequences," by Jeffrey N. Rosenthal, Timothy J. Miller, and Liam Leahy was published in The Legal Intelligencer on July 28, 2026. Reprinted with Permission.