Who Is Legally Liable When an AI Agent Makes a Mistake?
There is a meaningful difference between an AI tool that helps a human do something and an AI agent that goes out and does it on its own. That distinction is at the center of one of the most unsettled and consequential legal debates emerging in 2026.
Newly developed agentic AI systems do not simply generate text or produce a recommendation for a person to act on. They take real actions in the world. They send emails on your behalf. They execute financial transactions. They screen job applicants and reject them. They book travel, place purchase orders, and interact with third parties without waiting for a human to approve each step. The speed and autonomy that make these systems commercially attractive are the same properties that create serious legal exposure when something goes wrong.
Across industries, organizations are deploying AI agents faster than the legal frameworks governing them have developed. When an agent makes a costly mistake, however, affected parties are left asking a question that does not yet have a clean answer under U.S. law: who exactly is responsible?
If you were harmed by the actions of an autonomous AI system, whether you are a consumer, a job applicant, a business partner, or a patient, The Lyon Firm wants to hear from you. Contact us today for a free, confidential consultation.
What Makes Agentic AI Different From Other AI Tools?
Most people have some familiarity with AI tools that respond to prompts. You type a question, the system produces an answer, and you decide what to do with it. The human stays in the decision-making role throughout the interaction.
Agentic AI works differently. Once given a goal or a set of instructions, an agentic system figures out on its own how to accomplish that goal, taking whatever steps it determines are necessary. It can access external systems, interact with third parties, and carry out tasks that have real consequences, often with no human reviewing or approving individual actions along the way.
This shift from AI-as-tool to AI-as-actor is not a minor technical nuance. From a legal standpoint, it changes the entire analysis of fault, foreseeability, and accountability. A word processor that produces a bad draft causes no harm by itself. An AI agent that sends an unauthorized communication to a client, executes a transaction based on a flawed analysis, or locks a user out of a system on a false fraud determination creates actual, measurable harm, often before any human can intervene.
The Core Legal Problem: Accountability Without a Human Decision-Maker
Traditional liability law is built around human decision-making. In tort law, the central question is usually some version of: what did a reasonable person in this situation know, and what should they have done? In contract law, the question often turns on what the parties agreed to and whether someone breached that agreement. In product liability, courts look at the chain from manufacturer to consumer and ask where a defect originated.
The AI itself cannot be sued. It has no legal personhood, no assets, and cannot form the intent or mental state that criminal law typically requires. The harm is real, but the immediate actor, the system that took the action, exists outside the legal structures built to assign responsibility.
Who Can Be Held Legally Responsible for an AI Agent’s Actions?
When an agentic system causes harm, there are typically two categories of potential defendants worth examining: the developer who built the system and the deployer who put it into production.
The Developer
The company that designed, trained, and sold the AI agent is the first place courts and plaintiffs look for product liability claims. Under established U.S. product liability doctrine, parties harmed by a defective product can bring claims against any entity in the product’s supply chain. If the agent caused harm because of a flaw in its design or training, or because the developer failed to warn users of known limitations, a product liability claim may be viable.
The relevant defect theories in agentic AI cases include:
- Design defect: The system’s architecture or training methodology was fundamentally unsafe for the type of autonomous action it was authorized to take
- Manufacturing defect: The deployed version of the system deviated from the intended design in a way that caused harm
- Failure to warn: The developer did not adequately disclose to deployers or users the situations in which the agent was likely to fail or produce harmful outputs
The Deployer
The organization that chose to deploy the AI agent and granted it authority to act is often the more accessible defendant, especially for third parties who had no direct relationship with the developer. Under agency law principles, a business that authorizes another party to take actions on its behalf is generally responsible for the consequences of those actions when they fall within the scope of the authorized activity.
The law generally holds employers liable for the acts of employees acting within the scope of their employment. While courts have not uniformly applied this law to AI systems, the underlying logic is compelling: if a business deploys an agent with the authority to act on its behalf, it bears responsibility when that agent causes harm in the course of carrying out its assigned function.
A deployer’s liability exposure may be particularly strong in cases involving:
- Negligent deployment, meaning placing an AI agent into a high-stakes role without adequate testing, oversight, or safeguards
- Negligent supervision, meaning failing to monitor the agent’s outputs and catch harmful patterns before they caused widespread damage
- Breach of contract, when an AI agent acting on behalf of a vendor fails to meet contractual performance obligations or creates unauthorized obligations with third parties
Mobley v. Workday: What It Tells Us About Autonomous AI Decisions
One of the most significant cases in the emerging landscape of autonomous AI liability involves Derek Mobley, who applied to more than 100 positions using Workday’s AI-powered screening platform and was repeatedly rejected, sometimes within an hour of submitting an application, with no apparent human review. Mobley filed claims alleging that Workday’s algorithmic screening tool discriminated against him on the basis of race, age, and disability.
The case raises a question that will recur throughout AI agent litigation: when an autonomous system makes a decision that causes harm to a third party, does that third party have a viable claim against the organization that deployed the system, even if that organization’s human employees were not personally involved in the decision?
Courts are beginning to work through that question, and the direction of travel in the legal community suggests that deployers cannot insulate themselves from liability simply by pointing to the AI and arguing that the system acted on its own.
The “AI Did It” Defense Is Gone
California enacted legislation that took effect January 1, 2026, which directly addresses one of the most anticipated defenses in agentic AI litigation. Under that law, a defendant facing liability for harm caused by an AI system cannot use the system’s autonomous operation as a defense to the claim.
In plain terms: saying the AI made the decision does not get you off the hook. If your organization deployed an AI agent that caused harm, the fact that no human approved the specific action that caused that harm does not, by itself, defeat a liability claim.
This is a significant development. It forecloses what many observers expected to be a go-to strategy for defendants in AI liability cases and puts pressure on organizations to treat their deployed agents the way they would treat any employee or contractor acting with delegated authority.
Other states are moving in similar directions. Colorado’s AI Act, which takes effect in June 2026, requires deployers of high-risk AI systems to conduct regular impact assessments and maintain active risk management programs. New York City already requires annual bias audits for automated hiring tools. The regulatory landscape is tightening around autonomous AI, and with it, the legal liability for organizations that deploy these systems without adequate controls.
Real Situations Where Agentic AI Causes Legal Harm
To understand where these claims arise, it helps to look at the types of situations where agentic AI decisions cause real harm:
- An AI agent authorized to manage customer communications sends a message that makes representations your business cannot fulfill, and a customer relies on it to their detriment
- An automated fraud detection agent flags and suspends a legitimate account, locking out a customer who suffers measurable financial loss as a result
- An AI agent used in financial operations executes a transaction based on flawed analysis, causing investment losses the user never authorized under those conditions
- An AI hiring agent systematically rejects applicants from protected classes in ways that violate federal anti-discrimination law, with no human reviewer catching the pattern
- An AI agent used in a healthcare workflow generates a triage recommendation or scheduling decision that causes a patient to miss necessary care
In each of these situations, a human organization authorized the agent to act, the agent acted, and someone suffered harm. The legal question is not whether accountability exists but how to locate it and which legal theories best support a claim.
Why These Cases Are Legally Complex and Why That Matters
Agentic AI liability cases present challenges that most litigation does not. The systems are often technically opaque, meaning it can be genuinely difficult to reconstruct exactly why an agent made a particular decision.
That complexity does not mean injured parties lack options. It means they need attorneys willing to do the technical and legal work required to build a viable claim against well-resourced defendants.
The Lyon Firm has a long record of taking on complex, defendant-funded corporate litigation and pursuing accountability for individuals and businesses who were harmed by the decisions of large organizations. Attorney Joseph Lyon has served as lead counsel in state and federal class actions and has secured seven-figure results in cases involving defective products, corporate misconduct, and consumer harm. We handle agentic AI cases on a contingency basis, which means no fees or costs until we recover on your behalf.
If an autonomous AI system took an action that harmed you, your business, or your livelihood, you deserve to know what your legal options are. Contact The Lyon Firm today for a free and confidential consultation. We represent clients in Ohio, California, Illinois, Florida, and across the country.
Frequently Asked Questions
Can I sue a company for something an AI agent did without any human making the decision? Potentially, yes. Under emerging legal standards, including California law that took effect in 2026, organizations cannot use an AI system’s autonomous operation as a shield against liability. If you were harmed by an agent that a company deployed and authorized to act on its behalf, that company may bear legal responsibility.
What is the difference between an AI tool and an AI agent for legal purposes? An AI tool assists a human who retains decision-making authority. An AI agent takes actions on its own, interacting with the external world without requiring a human to approve each step. This distinction matters legally because it shifts the analysis from one involving human choices to one involving organizational decisions about deployment, oversight, and risk management.
Who should I sue if an AI agent caused me harm? That depends on the facts. In most situations, the most accessible defendant is the organization that deployed the agent and gave it authority to act. In cases where the harm stemmed from a fundamental defect in the AI system itself, claims may also run against the developer. A qualified attorney can help identify the right defendants and the strongest legal theories for your specific situation.