New York City puts companies on the hook for what their chatbots promise

The rule, in plain terms

New York City's proposed regulation does something simple: it says that when a business uses an AI chatbot to talk to customers, the business is responsible for what the chatbot says. If the chatbot promises a refund the terms don't allow, or gives wrong information about a product, that is the company's problem to fix ^[1].

The rule treats the chatbot as an agent of the business, in much the same way a shop assistant is. You cannot put up a sign saying "our staff might be wrong, check everything yourself" and walk away from what they tell customers. The same logic now applies to automation.

It is like putting your company logo on a letter. Once it goes out under your name, you own what it says, whether a person wrote it or a machine did.

What it covers

The regulation applies to businesses operating in New York City that use AI systems to interact with customers about products, services, returns, complaints or account matters ^[1]. It covers chatbots on websites, in apps and on messaging platforms.

It does not ban chatbots or require human review of every response. What it does is hold the company accountable for the consequences. If a chatbot misleads a customer and the customer acts on that information, the business has to deal with it as though a human employee had said the same thing ^[1].

That shifts the risk. Until now, many companies have treated chatbots as experimental: useful when they work, deniable when they do not. This rule says that is no longer an option.

How it differs from the AI Act

Article 50 of the AI Act requires businesses to tell people when they are interacting with an AI system, unless it is obvious from the circumstances ^[2]. The rule has applied across the EU since 2 August 2026. It is about transparency: people have a right to know they are talking to a machine.

The New York rule goes further. Transparency is assumed; accountability is the issue. Even if customers know they are talking to a chatbot, the company is still on the hook for what it says ^[1].

That matters because transparency alone does not settle disputes. Knowing you were talking to an AI does not tell you whether the company is bound by what the AI promised. The NYC rule answers that question: yes, it is.

The evidence gap

When a customer disputes what a chatbot said, the evidence question lands hard. In the 2024 Air Canada case, a tribunal in British Columbia ruled that the airline had to honour a bereavement fare the chatbot had wrongly described, even though the fare did not exist ^[3]. The customer had a screenshot. The airline offered no contrary evidence, and the tribunal noted that the absence of evidence suggested it would not have helped the airline's case.

That case turned on a screenshot because the airline had no record of its own. Under the New York rule, businesses face the same risk: if you cannot show what your chatbot actually said, the customer's version is likely to stand ^[1].

Screenshots are easy to take and hard to dispute when there is nothing to compare them with. A customer service team looking at a complaint six months later often has no way to check what the system said on a particular day, especially if conversations are not logged in full or if logs can be edited.

What a digital witness does here

This is a case where a digital witness fits well. The thing to seal is each answer the chatbot sends, or each conversation when it ends, together with the notice required under Article 50 that the customer was talking to an AI.

Sealing works by taking a fingerprint of the conversation record and sending it to WitniumChain. The content stays in your systems. To check later, you calculate the fingerprint again. If it matches, the record has not changed since it was sealed. Alter one word and it will not match.

Who can check? Customer service staff, the customer if you share the record, an ombudsman or a court. The check does not need our involvement and does not depend on trusting whoever keeps the log.

What it shows is that the record of what the chatbot said, and when, has not been edited. What it does not show is whether the chatbot's answer was correct, or whether the company is legally bound by it. Those are separate questions, but having an unchanged record is where the conversation starts.

How it would work

1. The chatbot sends an answer. Your system takes a fingerprint and seals it.
2. Months later, a customer disputes what was said. You retrieve the conversation and calculate the fingerprint again.
3. If it matches the sealed one, you can show the record has not changed. If the customer has a screenshot that contradicts your record, the sealed fingerprint tells you which version to trust.

You can seal answers as they happen or batch conversations every few minutes, depending on your setup. Start with the REST API, the SDK or the MCP server. The first 2,000 witnesses a month are free.

What this means for you

If you run customer-facing AI in New York City, or anywhere else where this kind of rule might follow, three questions are worth asking now:

1. Do we have our own record of what our chatbot said yesterday, last week, six months ago?
2. Can we show that record has not been edited, even by someone with administrator access?
3. If a customer turned up with a screenshot, could we check it against our own version?

The answers shape your risk. A business that can answer yes to all three is in a simpler position than one that cannot.

Beyond New York, the direction is clear. Regulators are done with the idea that a chatbot is a separate entity that might say anything. When it talks to your customers under your brand, it is you talking.

What to watch

The New York rule is still at the proposal stage. The city has invited public comment, and the final version may differ from what has been published ^[1]. Watch for the final text and the date it takes effect.

Outside the US, no equivalent rule has been proposed yet, but the logic is spreading. The AI Act's Article 26 requires deployers of high-risk AI systems to monitor the operation and to inform people affected by decisions ^[4]. That applies from 2 December 2027 for systems under Annex III. Accountability for what AI systems do is moving up the agenda.

Anyone building or buying AI for customer service should assume the direction is towards more accountability, not less, and that having your own checkable record will matter more as rules tighten.

Sources

1. Columbia Law School, Law and Political Economy Project, "NYC Proposes Rule Banning AI Chatbot Abuses", 2024, https://law-and-economy.law.columbia.edu/content/nyc-proposes-rule-banning-ai-chatbot-abuses
2. European Commission AI Act Service Desk, "Article 50: Transparency obligations for certain AI systems", https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-50
3. Civil Resolution Tribunal of British Columbia, "Moffatt v. Air Canada, 2024 BCCRT 149", 14 February 2024, https://decisions.civilresolutionbc.ca/crt/crtd/en/525448/1/document.do
4. European Commission AI Act Service Desk, "Article 26: Obligations of deployers of high-risk AI systems", https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-26