Attorney for Damage Caused by an AI System in New York City

AI systems now price inventory, screen job applicants, answer customer questions, approve and deny transactions, write marketing copy, and route calls for New York businesses of every size. When they work, nobody asks who is accountable for them. When they fail, the vendor points to its terms of service, the integrator points to the vendor, and the business that deployed the system points to the software. Meanwhile the person or company that was actually harmed is left with the loss.

"The AI did it" is not a legal defense. An AI system is a product, a service, or a tool that a person or company chose to build, sell, configure, and use. Existing New York law on contracts, warranties, negligence, deceptive practices, defamation, privacy, and discrimination applies to the people behind it. Our attorneys represent businesses and individuals in New York City who have suffered financial, reputational, or professional harm from an AI system, and we build the claim around the parties who had the power to prevent it.

Situations We Handle

AI harm cases rarely look alike, but most fall into a handful of patterns.

Failed AI deployments: A vendor sold your business a forecasting, underwriting, scheduling, or automation system that did not perform as promised, and you lost revenue, overpaid for inventory, or had to unwind operations built around it.
Chatbots and virtual agents that made promises: A company's customer-service bot quoted a price, a refund policy, or a coverage term, you relied on it, and the company now refuses to honor what its own system told you.
AI-generated falsehoods: A system published, or produced for others, false statements that you or your business committed misconduct, went bankrupt, or were sued, and customers or partners acted on them.
Automated decisions: An algorithm rejected your job application, cut your hours, froze your account, denied your credit, or terminated your seller or vendor account, with no human review and no explanation.
Data and privacy harm: An AI tool ingested your confidential business information, trade secrets, customer lists, or personal data and exposed or reused it.
Use of your name, voice, or likeness: A synthetic voice, image, or video used your identity in advertising or trade without your consent.
Operational damage: An AI agent with access to your systems deleted records, sent unauthorized communications, placed orders, or moved funds.

Who Can Be Held Responsible

Most AI harm involves a chain of companies. Identifying every link early matters, because contractual limitations, insurance, and the ability to pay a judgment are distributed very differently along it.

The developer of the model or system

The company that built and trained the system is responsible for defects in its design, its testing, and the warnings it gave about known limitations. Its liability usually runs through warranty, negligence, and, where the system can fairly be treated as a product, product liability theories.
The vendor, reseller, or integrator

Many businesses never contract with the model developer at all. They buy from a software company or consultant that wraps a third-party model in its own product and configures it. That company made its own representations in the sales process and in the statement of work, and it is often the most direct defendant. Our page on claims against IT contractors who fail to deliver covers the contract side of these relationships in more detail.
The business that deployed it

A company that puts an AI system in front of its customers, applicants, or the public is generally answerable for what that system says and does on its behalf, just as it would be for an employee or a printed brochure. In the widely reported 2024 Moffatt v. Air Canada decision, a Canadian tribunal rejected the airline's argument that its website chatbot was a separate entity responsible for its own statements. New York courts have not ruled on that exact question, but New York agency and misrepresentation principles point the same way.
The people who ignored the warning signs

Officers, managers, and professionals who approved a deployment despite known failures, or who relied on AI output in place of the professional judgment they were paid to exercise, may carry their own liability depending on their role and duties.

The vendor of a decision-making tool can also be a defendant in its own right. In Mobley v. Workday, Inc., a federal court in California allowed discrimination claims to proceed against the maker of an applicant-screening platform on the theory that it acted as the agent of the employers using it. That ruling does not bind New York courts, but it shows the direction of the argument.

Legal Theories Under New York Law

New York has no single statute that creates a cause of action for "AI damage." The claim is built from established law, and the strongest cases usually plead several theories at once.

Breach of contract: Service-level commitments, accuracy or performance specifications, acceptance criteria, data-handling terms, and statements of work are enforceable promises. Sales proposals and demonstrations are sometimes incorporated into the agreement. See our breach of contract page.
Breach of warranty: Where the AI system is sold as or with goods, UCC §§ 2-313 through 2-315 supply express warranties and implied warranties of merchantability and fitness for a particular purpose. Whether software is "goods" under Article 2 is unsettled in New York. Courts look at the predominant purpose of the transaction, and a subscription to a cloud service is more likely to be treated as a service contract, which leaves the claim to the contract's own warranty language. See breach of warranty.
Negligence and negligent misrepresentation: Available where the defendant owed a duty of care independent of the contract or, for misrepresentation, had a special relationship with the plaintiff. New York's economic loss rule (Bocre Leasing Corp. v. General Motors Corp., 84 N.Y.2d 685 (1995)) limits tort recovery for purely financial loss caused by a product, so how the harm is characterized matters.
Fraud: When a vendor knowingly overstated what its system could do, such as claiming human review that did not exist or accuracy figures it knew were false, a fraud claim can survive a general merger clause and a liability cap. A disclaimer that specifically names the representation at issue is harder to get past, so the exact contract wording matters. See business fraud.
Deceptive business practices: General Business Law §§ 349 and 350 prohibit consumer-oriented deception and false advertising. Under § 349(h) a plaintiff may recover actual damages or $50, whichever is greater, and up to $1,000 in treble damages for a willful or knowing violation. Under § 350-e the figures for false advertising are $500 and $10,000. Both sections allow an award of attorneys' fees.
Defamation: False statements of fact about a person or business, published to third parties, remain actionable whether a human or a model drafted them. Liability generally turns on who published the statement and what they knew. Claims against the AI developer itself are new and contested, including whether Section 230 of the Communications Decency Act protects output the system generated itself, so we often focus on the person or company that republished the falsehood. See business defamation.
Right of publicity: Civil Rights Law §§ 50 and 51 bar the use of a living person's name, portrait, picture, or voice for advertising or trade without written consent, and § 50-f protects deceased performers, including against digital replicas.
Employment discrimination: The New York State and New York City Human Rights Laws apply to decisions made with algorithmic tools exactly as they apply to decisions made by people. NYC Local Law 144 separately requires employers and employment agencies that use automated employment decision tools to obtain an annual independent bias audit, publish a summary of it, and give candidates advance notice. A missing audit or notice is useful evidence in a discrimination case.
Credit decisions: The Equal Credit Opportunity Act (15 U.S.C. § 1691(d)) and Regulation B require a creditor to give the specific reasons for denying credit. "The model decided" is not a specific reason, and the Fair Credit Reporting Act adds notice and dispute rights when a consumer report fed the decision.
Trade secret and confidentiality claims: Misappropriation, breach of a nondisclosure agreement, and breach of data-processing terms where a tool trained on or disclosed information it was not permitted to use. Related intellectual property questions are covered on our AI and intellectual property page.

The Contract Terms That Decide Most Vendor Cases

AI vendor agreements are written to shift risk onto the customer. Before any claim is filed, we review the terms that will shape it.

Clause	What it tries to do	Where it can fail
Limitation of liability	Caps damages, often at fees paid in the prior twelve months	New York will not enforce a cap against gross negligence or intentional wrongdoing (Sommer v. Federal Signal Corp., 79 N.Y.2d 540 (1992); Kalisch-Jarcho, Inc. v. City of New York, 58 N.Y.2d 377 (1983)), and a cap generally will not shield a party from its own fraud
Exclusion of consequential damages	Bars lost profits and business interruption losses	Lost profits that flow directly from the breach may be general damages that the exclusion does not reach (Biotronik A.G. v. Conor Medsystems Ireland, Ltd., 22 N.Y.3d 799 (2014))
"Outputs may be inaccurate" disclaimers	Shifts responsibility for errors to the user	Specific performance promises in an order form or statement of work usually control over boilerplate, unless an order-of-precedence clause says otherwise
Warranty disclaimer	Excludes implied warranties	UCC § 2-316 imposes conspicuousness and wording requirements that some clickthrough terms do not meet
Arbitration and forum clauses	Moves the dispute out of New York courts	Scope, notice, and whether the clause binds a party who never signed it are all open to challenge
Customer indemnity	Makes you pay for claims caused by the system	New York construes indemnity clauses strictly and will not stretch them past what the language clearly covers (Hooper Associates, Ltd. v. AGS Computers, Inc., 74 N.Y.2d 487 (1989))

Proving What the System Did

AI cases are won on records that disappear quickly. Vendors routinely purge prompt logs, conversation histories, and model versions on retention schedules measured in days or weeks, and a model that is retrained next month may no longer reproduce the behavior that harmed you.

Screenshots and exports of the outputs, with dates, times, and account identifiers
The prompts or inputs that produced them, if you have access
The contract, order form, statement of work, and every version of the terms you accepted
Sales decks, demonstrations, emails, and marketing claims made before you signed
Support tickets and the vendor's responses when you reported the problem
Records of the loss: invoices, lost orders, customer complaints, rejection notices

Once a dispute is reasonably anticipated, we send a written preservation demand to every company that holds relevant logs, model versions, and configuration records. The duty to preserve arises once litigation is reasonably anticipated; the letter removes any doubt about when that was, which is what supports spoliation sanctions if the evidence later goes missing. For companies that will not be parties, a subpoena may be needed. We then work with technical experts who can explain to a judge or arbitrator how the system reached its output and whether a reasonably designed and supervised system would have done the same.

Deadlines

The limitations period depends on the theory, and AI cases often combine several with very different clocks.

Defamation: one year from publication (CPLR 215(3)).
Negligence, property damage, and GBL § 349 claims: three years (CPLR 214).
Breach of warranty in a sale of goods: four years from tender of delivery (UCC § 2-725), unless the warranty explicitly extends to future performance.
Breach of contract: six years (CPLR 213(2)).
Fraud: the later of six years from the fraud or two years from when it was or reasonably could have been discovered (CPLR 213(8)).
Discrimination: court claims under the State and City Human Rights Laws generally have three years, but a federal charge must generally be filed with the EEOC within 300 days, and a complaint to the NYC Commission on Human Rights within one year for most claims. A missed agency deadline can end a federal claim.

Contracts can also shorten these periods or impose notice-of-claim requirements, and many AI vendor agreements do both.

What to Do Now

Capture the evidence: Screenshot and export everything before sessions expire or accounts are closed.
Stop the ongoing harm: Disable the system's access or suspend its use if you control it, and document when and why you did.
Do not sign anything new: Updated terms, credits, and "goodwill" offers from the vendor frequently contain releases.
Find your contract: Locate the signed agreement and any notice or dispute-resolution requirements in it.
Talk to counsel before notifying the vendor: The first written notice of a claim is also the first document the other side will quote back at you.

Related Practice Areas

Speak with a New York City Attorney About AI Harm

Send us a short description of what the system did, who provided it, and what it cost you, along with the contract if you have it. We will tell you who the likely defendants are, which deadlines apply, and what evidence needs to be preserved this week. We represent clients in Manhattan, Brooklyn, Queens, the Bronx, and Staten Island, in New York State courts, in the Southern and Eastern Districts of New York, and in arbitration.

You can contact us by phone at 212-233-1233 or by email at [email protected].