AI Agent Law: Authority, Liability and Personhood

1. Is software a legal agent?

No. Under U.S. law it is the instrument of the person who uses it.

Agency is a relationship between two persons: the principal assents to the agent acting on its behalf and under its control, and “the agent manifests assent or otherwise consents so to act.”² Software cannot consent. The Restatement reaches the same result: “a computer program is not capable of acting as a principal or an agent as defined by the common law. At present, computer programs are instrumentalities of the persons who use them.”³ If a program malfunctions, “even in ways unanticipated by its designer or user,” the user’s legal position is the same as with any other malfunctioning tool.⁴

The drafters of the Uniform Electronic Transactions Act, the basis of Florida’s e-commerce statute, took the same view: an electronic agent “is a tool of that person,” and “the employer of a tool is responsible for the results obtained by the use of that tool since the tool has no independent volition of its own.”⁵ They also looked ahead. They wrote that agents “may be created with the ability to act autonomously, and not just automatically,” and that courts “may construe the definition of electronic agent accordingly.”⁶ Today’s agentic systems test that prediction.

Florida’s agency tests assume people on both sides. Actual agency requires “(1) acknowledgment by the principal that the agent will act for him, (2) the agent’s acceptance of the undertaking, and (3) control by the principal over the actions of the agent.”⁷ Apparent agency requires a representation by the principal, reliance by a third party, and a change of position in reliance.⁸

What this means in practice. No U.S. court has held that an AI system is itself an agent. Courts reach the business instead. More than fifty years ago, the Tenth Circuit bound an insurer to its computer’s mistaken reinstatement of a policy: “A computer operates only in accordance with the information and directions supplied by its human programmers. If the computer does not think like a man, it is man’s fault.”⁹ More recently, a federal court let discrimination claims proceed against an AI hiring vendor as an “agent” of the employers that used it. The agent there was the vendor company, not its software: “Employers cannot escape liability for discrimination by delegating their traditional functions, like hiring, to a third party.”¹⁰

“However advanced the Assistant currently is, it is a tool, not a person for statutory purposes.”

Ninth Circuit, Amazon v. Perplexity (2026)

2. Contracts formed by AI agents

A contract an AI agent forms is generally valid and binds the business that deployed it.

Federal law provides that a contract “may not be denied legal effect, validity, or enforceability solely because its formation … involved the action of one or more electronic agents so long as the action of any such electronic agent is legally attributable to the person to be bound.”¹¹ Florida goes further: “A contract may be formed by the interaction of electronic agents of the parties, even if no individual was aware of or reviewed the electronic agents’ actions or the resulting terms and agreements.”¹² Florida defines an “electronic agent” as “a computer program or an electronic or other automated means used independently to initiate an action or respond to electronic records or performances in whole or in part, without review or action by an individual.”¹³

Whose act is it? A record is attributable to a person “if the record or signature was the act of the person,” which may be shown “in any manner.”¹⁴ The uniform comments say an agent’s act is the deployer’s act, so that the record “is not ascribed to a machine, as opposed to the person operating or programing the machine.”¹⁵ Intent comes from deployment: “When machines are involved, the requisite intention flows from the programing and use of the machine.”¹⁶

When something goes wrong. Florida’s statutory error remedy is narrow. It protects an individual who makes a mistake while dealing with someone else’s agent, and only if that agent gave no chance to prevent or correct the error and the individual promptly gives notice, returns what was received and takes no benefit.¹⁷ Businesses cannot contract around it.¹⁸ Errors the agent makes itself are treated differently: “If the error results from the electronic agent, it would constitute a system error,” governed by the ordinary law of mistake.¹⁹ That law does not always hold a business to its own error. In Florida, a contract may be set aside for a unilateral mistake of material fact if the mistake was not the result of an inexcusable lack of due care, denying relief would be inequitable, and the other party has not changed its position in reliance so far that rescission would be unconscionable. The mistaken party does not have to show that the other side induced the error.²⁰ The test turns on the facts, so a business should not count on it. The uniform comments also point out that confirmation screens give a business a practical way to manage error risk.²¹

Assent still matters. An automated, ministerial acknowledgment is not an acceptance.²² Florida courts enforce browsewrap terms only where the user actually knew of them or the notice is conspicuous.²³

Open question. No reported U.S. decision has let a business avoid a contract on the ground that its AI agent exceeded instructions. Attribution rules suggest the deployer will usually be held to the deal.

“When machines are involved, the requisite intention flows from the programing and use of the machine.”

Uniform Electronic Transactions Act § 14, official comment

3. Authority and its limits

Expect to be held to what your agent presents to the world, although no reported U.S. decision has yet applied apparent authority to an AI agent.

Because software is not a legal agent, the familiar authority doctrines apply to the business that deploys it. In Florida, apparent agency “exists only if each of three elements are present: (a) a representation by the purported principal; (b) a reliance on that representation by a third party; and (c) a change in position by the third party in reliance on the representation.”²⁴ When a company builds, brands and deploys a customer-facing agent, the company makes the representation.

Ratification requires more. Florida requires that the principal be “fully informed” and approve the act; constructive knowledge is generally not enough.²⁵ Unlogged agent activity, therefore, may not count as ratification. But it may still be attributed to the company under the electronic-transactions rules discussed above.

The most-cited case on a chatbot’s statements is Canadian. A British Columbia tribunal held an airline liable for its website chatbot’s wrong answer about bereavement fares. The airline argued, in effect, that the chatbot was “a separate legal entity that is responsible for its own actions.” The tribunal called that “a remarkable submission,” noting that a chatbot “is still just a part of Air Canada’s website.”²⁶ The case was decided as negligent misrepresentation, it is persuasive only, and no U.S. appellate court has yet applied apparent authority to an AI agent.

Practical limits that hold up. Narrow credentials and spending caps, human approval for defined categories of transactions, clear disclosure of what the agent can and cannot commit to, confirmation screens, and contract terms that allocate agent-error risk with vendors and counterparties.

4. Liability when the agent gets it wrong

Expect courts to look past the software to the developer and the deployer.

The “AI did it” defense. No court has accepted it. California bars it by statute: for anyone who “developed, modified, or used” AI alleged to have caused harm, “it shall not be a defense, and the defendant may not assert, that the artificial intelligence autonomously caused the harm to the plaintiff.”²⁷ Causation, foreseeability and comparative-fault arguments remain available.²⁸ Utah bars a similar defense in consumer-protection enforcement.²⁹ Florida has no such statute, so in Florida the question is argued as ordinary causation and foreseeability.

Products liability — the Florida cases. A Palm Beach County court concluded that the Lyft app was a “product” under Florida product liability law because the claims arose from defects in the app, not from its ideas or expression.³⁰ Applying that reasoning, the Middle District of Florida allowed product claims against an AI companion-chatbot developer to proceed at the pleading stage, and let claims proceed against Google as an alleged component-part maker and aider and abettor.³¹ The court also declined, at that stage, to hold that the chatbot’s output was protected speech. The case later settled.³² In Miami, a federal jury found Tesla 33% at fault for a fatal crash involving its Autopilot driver-assistance system. The court entered a judgment of about $242.6 million and denied post-trial relief; an Eleventh Circuit appeal is pending.³³

Not every court agrees. Other courts hold that information, ideas and algorithms are not “products.”³⁴

Section 230. Whether the federal internet-immunity statute protects text an AI model writes itself is unsettled. One federal court held that using generative AI for content moderation does not forfeit immunity.³⁵ Another held that a platform’s generative-AI ad tool that allegedly rewrote fraudulent ads raised a fact question that defeated immunity at the pleading stage.³⁶ No appellate court has decided the issue.

Defamation by “hallucination.” A Georgia court granted summary judgment to OpenAI on a defamation claim over a fabricated ChatGPT output. It found that no reasonable reader in the user’s position would take the output as fact, and that the plaintiff could show neither fault nor damages.³⁷

Consumer protection. The Florida chatbot case also sustained a Florida Deceptive and Unfair Trade Practices Act claim at the pleading stage.³⁸

“It should be obvious to Air Canada that it is responsible for all the information on its website. It makes no difference whether the information comes from a static page or a chatbot.”

B.C. Civil Resolution Tribunal, Moffatt v. Air Canada (2024)

5. Agents on other people’s platforms

When an agent runs in the user’s own session at the user’s direction, the user is the one accessing the site, at least under federal anti-hacking law as the Ninth Circuit reads it.

The federal Computer Fraud and Abuse Act turns on “access.” The Supreme Court reads “exceeds authorized access” as “a gates-up-or-down inquiry.” Either one “can or cannot access a computer system,” or certain areas within it.³⁹ The Court left open whether the gates must be technical or may also be set by contract or policy.⁴⁰

In Amazon v. Perplexity, the Ninth Circuit vacated a preliminary injunction that would have barred an AI shopping agent from Amazon accounts. The court concluded: “It is the user who ‘accesses’ Amazon’s computers, with the help of the Assistant to carry out specific acts on Amazon.com.”⁴¹ It drew limits. It did “not establish a new legal regime governing agentic AI.” It did “not address whether in other contexts, including tort claims,” the developer “can avoid liability for the Assistant’s actions.”⁴² And it said platforms keep the ability “to regulate access … via private terms of service for its users.”⁴³ An earlier Ninth Circuit case remains relevant. There, a cease-and-desist letter revoked the permission users had given a third-party service, and continued access became unauthorized.⁴⁴

The Florida and Eleventh Circuit picture. The Eleventh Circuit’s older purpose-based reading of the statute has been abrogated by the Supreme Court.⁴⁵ No published Eleventh Circuit decision yet applies the current test to AI agents. Florida’s criminal computer statute prohibits access “without authorization or exceeding authorization” but does not define “authorization.”⁴⁶ Florida’s civil Computer Abuse and Data Recovery Act is built around “technological access barrier[s]”: passwords, security codes, tokens and similar measures. It reaches access, with intent to cause harm or loss, by someone who is not an authorized user, who has stolen a barrier, or who circumvents one without the owner’s permission.⁴⁷ Whether a customer’s agent using the customer’s own password “circumvents” anything is untested.

Bottom line. For platforms, terms of service and technical controls are the main tools. For agent builders, design so that the user, not the builder, stays in control, and expect contract and tort claims even where anti-hacking claims fail.

“We do not establish a new legal regime governing agentic AI.”

Ninth Circuit, Amazon v. Perplexity (2026)

6. Personhood and entity wrappers

No U.S. court or statute recognizes an AI system as a legal person, and several states have legislated the opposite.

Courts tie legal status to human beings or to entities through which humans act. Only “natural persons” can be patent inventors.⁴⁸ Copyright requires work “authored in the first instance by a human being.”⁴⁹ Animals have no standing to sue under a statute unless the statute plainly grants it.⁵⁰ New York’s highest court linked personhood to “the capacity, not just to benefit from the provision of legal rights, but also to assume legal duties and social responsibilities.”⁵¹

State statutes (as of October 2026).

• Idaho prohibits it: AI “shall not be granted personhood in the state of Idaho.”⁵²
• Utah bars governmental entities from granting or recognizing legal personhood in AI.⁵³
• North Dakota defines “person” to exclude AI. This is a definition, not a prohibition.⁵⁴
• Tennessee (2026) likewise defines “person” to exclude “artificial intelligence, a computer algorithm, a software program, computer hardware, or any type of machine.”⁵⁵
• Oklahoma HB 3546 passed the House but died in the Senate. Ohio HB 469 remains in committee.⁵⁶

Florida. Florida has no AI-personhood statute. Its general definition of “person” is broad and says nothing about AI.⁵⁷ The 2026 “AI Bill of Rights” (SB 482, then SB 2-D in special session) passed the Senate twice but died in the House.⁵⁸

Entity wrappers. Scholars have argued that flexible LLCs can act as legal “containers” for autonomous systems.⁵⁹ Wyoming allows a DAO LLC’s management to be “vested in its members or the members and any applicable smart contracts,” and requires the articles to say how far management is “conducted algorithmically.”⁶⁰ Florida is less accommodating. An LLC “manager” must be a “person,”⁶¹ and an LLC dissolves after “90 consecutive days during which the company has no members” unless the transferees cure it.⁶² A software-run Florida LLC therefore still needs a human or entity behind it.

Skipping the wrapper is riskier. Courts have treated unwrapped DAOs as unincorporated associations that can be sued and held liable,⁶³ and their governance participants, including venture investors, as plausible general partners with personal exposure.⁶⁴

7. Duties and conflicts

The AI owes no duties. The people and firms who deploy it do.

In Florida, a fiduciary claim requires “the existence of a fiduciary duty, and the breach of that duty such that it is the proximate cause of the plaintiff’s damages.”⁶⁵ Agents owe loyalty. They must “remain loyal to the interests of [the] principal during the continuation of [the] agency.”⁶⁶ When an AI agent acts for a client, these duties sit with the provider and the deployer.

Regulated advisers. Investment advisers are fiduciaries whose advice must be free of undisclosed conflicts.⁶⁷ Automating that advice does not change the duty.

Disclosure. California makes it unlawful to use an undisclosed bot “with the intent to mislead the other person about its artificial identity” to drive a sale or influence a vote.⁶⁸ Utah requires businesses to disclose generative-AI use when a consumer clearly asks, and provides a safe harbor for clear disclosure at the outset of and throughout the interaction.⁶⁹ Florida has no general AI-disclosure statute, but deceptive AI claims fall under the Florida Deceptive and Unfair Trade Practices Act and the FTC Act.

For Florida lawyers. The Florida Bar’s ethics opinion on generative AI requires client-facing chatbots to “include a disclaimer indicating that the chatbot is an AI program and not a lawyer or employee of the law firm,” and requires lawyers to “verify the accuracy and sufficiency of all research performed by generative AI.”⁷⁰ Courts in the Eleventh Circuit enforce those duties. One court sanctioned a lawyer’s unverified AI citations as “recklessness in the extreme, and … tantamount to bad faith.”⁷¹ Another dismissed a pro se plaintiff’s case, in part as a sanction for AI-fabricated citations, noting the “duty of candor to the Court.”⁷²

Key authorities

State AI statute tracker

Current through October 6, 2026

Frequently asked questions

Is an AI agent a legal “agent” under U.S. law?

No. Under the Restatement (Third) of Agency, a computer program cannot be a principal or an agent. It is an instrument of the person who uses it. The business that deploys the AI is responsible for what it does.

If my AI agent signs a contract, am I bound?

Generally, yes. Federal law and Florida’s Uniform Electronic Transaction Act allow contracts to be formed by electronic agents even when no person reviews the terms, as long as the agent’s actions are attributable to you.

Can I undo a contract if my AI agent made a mistake?

Possibly, but don’t count on it. Florida’s statutory error remedy protects individuals who make mistakes while dealing with someone else’s automated system. An error by your own agent is handled under the ordinary law of mistake. Florida courts can set a contract aside for a unilateral mistake, but only on a fact-specific showing that includes no inexcusable lack of due care.

Can I argue “the AI did it” to avoid liability?

In California, no: a statute in effect since January 1, 2026 bars that defense. No court has accepted it elsewhere. Florida has no similar statute, so in Florida the question turns on ordinary causation and foreseeability.

Is it illegal for an AI agent to log into a website using my password?

Under the Ninth Circuit’s 2026 Amazon v. Perplexity decision, when the agent works in your session at your direction, you, not the agent’s maker, are the one accessing the site for federal anti-hacking purposes. Platforms can still restrict access through their terms of service, and the Eleventh Circuit, which covers Florida, has not yet ruled on AI agents.

Can an AI be a legal person or own a company?

No U.S. jurisdiction recognizes AI as a legal person. Idaho and Utah prohibit it, and North Dakota and Tennessee define “person” to exclude it. Some states, like Wyoming, allow entities to be managed partly by smart contracts, but a human or legal entity still stands behind them. In Florida, an LLC manager must be a “person.”

Does Florida have an AI disclosure law?

Not a general one. Florida’s 2026 “AI Bill of Rights” did not pass. Deceptive AI practices can still violate Florida’s Deceptive and Unfair Trade Practices Act. Florida lawyers who use client-facing chatbots must disclose that the chatbot is AI under Florida Bar Ethics Opinion 24-1.