Your Chatbot Does Not Need Authority to Create an Obligation

A law firm's chatbot advised a potential client that they could take a certain legal action. The firm later said the advice was wrong and that the chatbot lacked authority to give it. The client sued anyway. The case turned on a simple question: did the chatbot's statement create an obligation, or was it just a machine talking out of turn?

Jones Walker LLP's AI Law Blog looked at the question in November 2024 and reached a conclusion worth remembering: your chatbot does not need formal authority to create problems ^[1]. Think of it like this: a hotel receptionist cannot change the room rate in the system, but if they quote the wrong price to a guest, the hotel still has a dispute to sort out. The chatbot is the same. It can mislead, and it can create mess, even when it has no power to rewrite your contracts.

What the law firm case shows

The law firm argued that the chatbot's advice did not bind it because the chatbot was not authorised to give legal advice or to enter into contracts on behalf of the firm ^[1]. That sounds logical, but the court was not convinced. The question, the blog explains, was not whether the chatbot had authority in some abstract sense, but whether a reasonable person relying on the statement could suffer harm or be misled ^[1].

The case raises two points. One, the chatbot's lack of formal authority did not shield the firm from liability for a misleading statement. Two, the firm's argument that the chatbot was "a separate legal entity" responsible for its own actions was, to put it kindly, creative ^[1]. A chatbot is not a separate legal entity any more than a vending machine is. It is a tool, and the organisation that deploys it is responsible for what it does.

Article 50 and the transparency rule

Under Article 50 of the AI Act, people must as a rule be told when they are interacting with an AI system, from 2 August 2026 ^[2]. The rule is simple: no hiding. If your chatbot is answering questions, customers must know they are not talking to a person.

The requirement applies to AI systems designed to interact with natural persons ^[2]. There are narrow exceptions, for example when it is obvious from the circumstances or when the law allows it for detecting or preventing criminal offences, but in customer service the default is disclosure ^[2].

That transparency requirement does not settle liability. It just means customers will know they are talking to a machine. Whether that machine's statements bind you is another matter, and the law firm case suggests that formal authority is not the deciding factor.

The evidence question: whose record wins?

When a customer disputes what a chatbot said, the evidence is often a screenshot. The customer has one. You might have server logs, but those logs are kept by you, and a sceptical counterparty or a court might wonder whether they have been tidied up since the conversation.

That is the evidence question, and it comes up every time a chatbot makes a promise, gives advice or quotes a price. What did it actually say, and when? If you cannot show that in a way the other side can check, you start the dispute on the back foot.

How a digital witness fits here

The thing to seal is each answer the chatbot sends, or each conversation when it closes. That includes the notice that the customer was talking to an AI, as Article 50 requires. Sealing takes a fingerprint of the answer and the timestamp, not the content itself, so nothing private leaves your systems.

When the answer is sent, your system calls WitniumChain through the REST API, the SDK or the MCP server and gets a witness ID back. The record stays with you. If it is questioned later, anyone you share it with can calculate the fingerprint again and check it against the sealed one. If it matches, the record has not changed since it was sent. Change a single character and it will not match.

That gives you your own version of the screenshot: a record of what the chatbot said, when it said it and what the customer was told about talking to an AI, in a form nobody can quietly edit afterwards. It does not settle whether the firm is bound by the statement, but it stops the dispute being a question of whose memory is better. The first 2,000 witnesses a month are free.

How it would work

1. Customer asks the chatbot a question.
2. Chatbot generates an answer, your system seals it and sends both the answer and the Article 50 notice to the customer.
3. If the answer is disputed, you produce the sealed record and invite the other side to check the fingerprint.

What this means for you

If your organisation uses a chatbot to talk to customers, ask yourself three questions. First, could you show what the chatbot said on a particular day, in a form the other side can verify? Second, is the Article 50 notice included in the record you keep? Third, if the chatbot gives wrong advice, do you have a process for catching and correcting it before it creates a liability?

The law firm case is a reminder that formal authority is not a shield. A chatbot can create trouble even when it has no power to act on your behalf. The answer is not to shut the chatbot down, but to keep a checkable record of what it says and to make sure someone is watching.

What to watch

Article 50 of the AI Act applies from 2 August 2026 across the EU. If you are deploying or updating customer-facing AI before that date, build the notice into the design now. Non-compliance with Article 50 can attract fines of up to €15 million or 3 per cent of global annual turnover, whichever is higher ^[2].

Sources

1. Jones Walker LLP, AI Law Blog, "Your Chatbot Does Not Need Authority to Create an Obligation", November 2024, https://www.joneswalker.com/en/insights/blogs/ai-law-blog/your-chatbot-does-not-need-authority-to-create-an-obligation.html
2. European Commission AI Act Service Desk, Article 50, Regulation (EU) 2024/1689, https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-50