Can a chatbot *provider* be liable for publishing your article?

The opinion in brief

On a date not yet public, Advocate General Maciej Szpunar delivered an opinion in a case before the Court of Justice of the European Union ^[1]. The case concerns whether the provider of an online chatbot can be held liable for copyright infringement when the chatbot reproduces press publications without the publisher's consent.

The Advocate General's answer is yes, in two situations: when the use of press content is inherent to the provider's business model, or when the provider has failed to implement credible and effective measures to prevent infringement.

It's like running a photocopier in a library. If you put the machine there, charge for copies and don't put up a sign saying "check the copyright first", you can't later say the person pressing the button is solely to blame.

What the law says

Under the Copyright Directive, press publishers have the right to authorise or prohibit the online use of their publications by information society service providers ^[2]. That right lasts for two years from publication.

A chatbot that scrapes articles, stores them and reproduces parts of them in answers is using those publications. The question the Court is considering is whether the provider of the chatbot, not just the person who asked the question, can be held liable for that use.

The Advocate General's reasoning

The opinion sets out two grounds for provider liability.

First, when the use of press content is inherent to the business model. If scraping and reproducing articles is what makes the service work, the provider is responsible for that act of use.

Second, even if the provider argues it is merely an intermediary, it can still be liable if it has not taken credible and effective measures to prevent infringement. The opinion suggests that simply saying "we are not responsible" is not enough. The provider must actively try to stop unlawful use.

Why it matters

This opinion, if followed by the Court, would make it harder for chatbot providers to argue that they are neutral platforms with no control over what their systems do. It would align AI providers' responsibilities with those of other online service providers, such as hosting platforms and search engines, which have long been expected to take steps against unlawful content.

It also fits with the transparency duties in the AI Act. Article 50 already says that people must be told when they are talking to an AI, and that AI-generated content must be marked ^[3]. This opinion goes further: it says the provider is responsible for what the AI does with other people's work.

The path to a final judgment

An Advocate General's opinion is not binding, but the Court follows such opinions in around 70 to 80 per cent of cases ^[4]. The Court will now consider the arguments and deliver its judgment, which could take several months.

Press publishers, AI developers and anyone building chatbots that use third-party content will be watching.

What this means for you

If you are building or running a chatbot that uses content from the web, this opinion suggests three questions to ask now:

1. Does your service scrape or reproduce press publications as part of its core function? If so, do you have licences or another legal basis?
2. Have you implemented measures to prevent unlawful use of protected content? Can you show what those measures are?
3. If a publisher complained that your chatbot reproduced their article without permission, could you show exactly what your system retrieved, stored and used?

The opinion doesn't say you need a digital witness, but it does say you need to take responsibility for what your AI does. That starts with knowing what it did.

What a witness would show, and what it wouldn't

Here, the thing to seal is each retrieval and each answer: what content was fetched, from where, when it was stored and what was sent to the user. Sealing takes a fingerprint, not the content itself, so you can seal a record of a copyrighted article without storing or transmitting the article through a third party.

Anyone you share that record with can check that it hasn't changed since it was sealed, without having to trust your word. That includes a publisher who claims you used their work, a supervisor or a court.

The witness shows that the record is the one created at the time. It doesn't show that the use was lawful, that a licence existed or that your countermeasures were adequate. Those are separate questions, but they start with knowing what happened.

A digital witness per retrieval and per answer makes that a short conversation rather than a long one. You seal the record as each event happens through WitniumChain, using the REST API, the SDK or the MCP server, and the content stays with you. The first 2,000 witnesses a month are free. Read the docs.

How it would work

1. Chatbot fetches an article: seal a record of the URL, timestamp and content fingerprint.
2. Chatbot generates an answer: seal the question, the answer and the sources used.
3. Publisher questions the use: you share the sealed record, they calculate the fingerprint and see it matches.

What to watch

The Court's final judgment, expected in the coming months, will settle whether this reasoning becomes binding law. In the meantime, the AI Act's transparency requirements under Article 50 apply from 2 August 2026, and the broader high-risk rules, including logging duties under Article 12, apply to Annex III systems from 2 December 2027.

If you run a chatbot that uses third-party content, the prudent step is to act as if the opinion will be followed.