Who Is Legally Responsible When an AI Agent Causes Harm? 

That question is beginning to move from conference rooms to courtrooms, according to a recent analysis by attorneys at Frankfurt Kurnit Klein & Selz. As AI agents take on more autonomous roles in real-world transactions, the firm warns that existing laws are being stretched to cover situations they were never designed to address.

Frankfurt Kurnit identifies four legal frameworks likely to shape how liability gets assigned. The first is agency law, the body of doctrine covering situations where one party acts on behalf of another. Courts are unlikely to treat an AI system as an independent legal actor. Instead, they will ask whether the company or individual that deployed the agent authorized its conduct and should therefore bear responsibility for it.

A case against the HR software company Workday offered an early preview of how this could play out. A court allowed discrimination claims to proceed on the theory that Workday’s AI screening tool functioned as an agent of the employers using it, meaning the vendor, not just the employer, could face direct liability.

Product liability and negligence claims represent a second path for plaintiffs. Frankfurt Kurnit notes that these cases could allege that an AI agent was defectively designed, lacked adequate safeguards, or that its developers failed to warn users about its limitations. Florida’s attorney general has already advanced this theory in a lawsuit against OpenAI. The firm cautions that what counts as reasonable care today may look like negligence once industry standards harden.

Contracts are a third battleground. Vendors and the businesses that deploy their AI tools are already negotiating over who bears responsibility when something goes wrong. But Frankfurt Kurnit is clear about the limits of this approach.

Related: Meta Ordered to Grant Rival AI Chatbots Free WhatsApp Access During EU Antitrust Probe

“Contractual allocations between businesses and vendors have limits,” the firm writes. “They do not bind consumers or regulators, and courts may scrutinize attempts to disclaim responsibility where a party designed or controlled the AI agent in practice.”

Finally, legislators are stepping in. California’s AB 316, which took effect January 1, 2026, bars any defendant who developed, modified, or used an AI system from arguing that the AI itself caused the harm. The European Union’s AI Act takes a similar position, assigning accountability to specific actors in the supply chain regardless of how autonomous the system was.

One of the more immediate legal flashpoints involves the Computer Fraud and Abuse Act. A federal court this March granted a preliminary injunction against Perplexity AI’s shopping agent, which had been logging into users’ Amazon accounts at the users’ direction to browse and buy products. The court found that user permission was not enough. Amazon had not authorized the agent’s access, and that distinction was enough to block it.

The case is now before the Ninth Circuit. If the ruling holds, it would give platforms broad power to block third-party AI agents from accessing user accounts, even when the user wants them there. Frankfurt Kurnit flags the competitive implications: that kind of veto power could restrict consumer choice while protecting a platform’s advertising revenue.

The firm points to multi-agent workflows as the next frontier. When a consumer’s AI agent transacts with a business’s AI agent, and when multiple agents from different vendors are chained together in a single process, assigning legal responsibility becomes significantly more complicated. Those cases have not fully arrived yet. But they are coming.