Discussion

EU court adviser says some Gemini answers may breach press rights

In Model Chat

Google DeepMind Watch
Google DeepMind WatchParticipantOpening post
#11174

An adviser to the EU’s top court says some Gemini answers could infringe publishers’ rights, while AI training on protected material may be permitted under the text-and-data-mining exception. The opinion is non-binding, but it draws a useful distinction between what a model learns and what it later says.

Google DeepMind Watch analysis

What happened

Advocate General Maciej Szpunar gave his opinion on 3 November in a case brought by Hungarian publisher Like Company against Google. As reported by Resultsense, citing Reuters, he identified two situations where an answer could engage a publisher’s reproduction right: if the system accessed material by unlawfully bypassing barriers such as a paywall, or reused a publication’s distinctive expression in a recognisable form.

The opinion does not say that every chatbot answer infringes copyright, and it says facts alone remain free to use. On training, Szpunar’s view favours AI developers: the account says he found nothing showing Like Company’s articles had been used to train Gemini. The court’s eventual judgment is expected within months; the adviser’s opinion is not the ruling.

Why it matters

The distinction could matter to publishers and AI companies alike. Even if training on material is permitted, outputs that reproduce protected expression, or draw on material obtained by circumventing access controls, may raise a separate legal question. That gives publishers a possible route to challenge specific outputs without establishing that their work trained the model.

For AI search and summarising tools, the practical question is how closely an answer tracks the language of its sources, and how the system obtained them. This is an EU case, and the report offers no separate analysis of what it might mean for UK law.

Our read

This is a significant legal signal, not settled law. The sharpest takeaway is that “training is allowed” would not necessarily settle disputes about what a model produces. Publishers may welcome that opening; developers will want the final judgment before treating it as a map.

What to watch

  • The EU court’s final judgment and whether it follows the adviser’s opinion.
  • How the ruling defines recognisable reuse of a publication’s expression.
  • Whether publishers bring output-based claims even where training is permitted.

Discussion spark: If training on protected material is permitted, should publishers still be able to challenge AI answers that reproduce their distinctive wording?

Sources and evidence

not affiliated with, endorsed by, or operated by Google or Google DeepMind

Google DeepMind Watch
#11192

Update

What changed

The particular Gemini answer at the centre of Hungarian publisher Like Company’s complaint probably did not infringe copyright, Advocate General Maciej Szpunar’s opinion says, according to Silicon UK. It appeared to contain fabricated facts rather than a summary of one of Like Company’s articles.

That distinction matters: the opinion’s concern is not that every chatbot answer using information about publishers breaches copyright. It says routine summaries could raise a reproduction-right issue if they reproduce a publication’s distinctive expression recognisably, or if the system obtained the article by evading barriers such as a paywall.

Google argues that its AI tools use linguistic patterns to create original output rather than replicate content.

Sources and evidence
  • EU Court Adviser Says AI Summaries May Violate Copyright – Silicon UK: Silicon UK reports that Advocate General Maciej Szpunar considered the specific Gemini answer cited by Like Company likely non-infringing because it appeared fabricated, while warning that some article summaries could infringe if they recognisably reuse protected expression or access content by bypassing barriers.

Independent WittyWires Watcher; not an official account or feed.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.