Meta wants to read company websites. Is it possible to refuse AI training?
Polish publishers are asking UOKiK to block Meta’s new terms, which from November 20 will allow it to copy content from company websites. In Norway, a court has already suspended them, and Google, Microsoft and OpenAI are announcing a joint rule that a website can use to exclude its content from AI training.
From November 20, companies using Meta’s business services are expected to accept new commercial terms. The biggest objection concerns the so-called crawling clause. It gives the group consent for its automated programs to search, copy, store and analyze the contents of business users’ websites. The Polish Chamber of Press Publishers (PIW) and Bonnier Business Polska, the publisher of “Puls Biznesu,” considered this an abuse of dominant position and notified the Office of Competition and Consumer Protection (UOKiK).
For now, the dispute involves media companies, but it concerns every company that has a Facebook or Instagram profile or buys ads there. What is at stake is who decides whether content from a company website ends up in artificial intelligence systems.
What exactly Meta’s new terms change
The new rules were supposed to apply from October 30, but Meta moved the date to November 20. According to publishers, the clause allows the use of, among other things, scrapers and bots, meaning programs that automatically download website content. It also covers the creation of test login credentials. Publishers warn that this opens the way to bypassing registration and paywalls (paid barriers) and accessing content available only to subscribers.
The first version of the terms explicitly referred to training models. Around October 16, Meta removed the word “train,” and the current wording refers to using content to develop, improve, research and maintain advertising and business products. The European Publishers Council, an organization representing European publishers, considers this change cosmetic: the new wording does not explicitly exclude AI training or the generation of answers and summaries.
Meta disagrees. Its spokesperson Daniel Roberts said the disputed provisions concern only advertising and commercial products, and that website owners can still block the group’s crawlers in the robots.txt file (a file on the server in which a website tells bots what they are not allowed to download).
Can a company refuse
In short: it can, but at the cost of giving up the services. According to PIW, agreeing to the clause is a condition for continued use of Facebook and Instagram pages as well as Meta’s advertising and analytics tools. It is not possible to reject just the clause and keep the rest. Using advertising products after November 20 will mean accepting the new terms.
The robots.txt entry that Meta refers to is a technical signal, not an opt-out from the contract. Publishers point out that from the outside it is difficult to determine which bot is acting on whose behalf and for what purpose. There is also copyright law: PIW believes the new terms weaken press publishers’ neighboring rights and the TDM opt-out mechanism, that is, an objection to text and data mining. According to the chamber, once content is included in Meta’s AI systems, it may in practice be impossible to identify and remove it.
Oslo stopped Meta, Warsaw is waiting for the authority
In letters to UOKiK president Tomasz Chróstny, PIW and Bonnier cite Article 102 of the Treaty on the Functioning of the European Union, the Polish Act on Competition and Consumer Protection, and the Digital Markets Act (DMA), which imposes on the largest platforms an obligation to offer fair and non-discriminatory terms to business users. They are requesting interim measures prohibiting the use of the clause against publishers in Poland, as well as informing the European Commission about the matter and assessing whether the proceedings should not take place at the EU level. For now, this is a notification and a motion; there is no information that the authority has opened proceedings.
Norway is one step further. On October 28, a court in Oslo issued an interim ban on introducing the new terms. The motion was filed by the Norwegian media companies’ association MBL on behalf of ten companies, including public broadcaster NRK, Schibsted, the NTB agency, and the Amedia, Polaris and Aller groups. At the preliminary stage, the court found that Meta may be abusing its dominant position and that publishers risk irreversibly losing control over journalism. The ruling was issued without hearing Meta, which may request a hearing with both parties present. This is an interim ban that is to remain in force until a final decision, not a judgment on the merits.
A day later, on October 29, the UK competition authority CMA opened an investigation into Meta. If an infringement is confirmed, the penalty could be up to 10 percent of global turnover. Meta responds that there are no grounds for proceedings. Criticism is also coming from news agencies associated in MINDS International:
“It is unacceptable for a global technology company to use its market position to demand rights to journalistic content that it neither created nor paid for.”
A joint AI training block in robots.txt: a promise for 2027
At the same time, Meta’s competitors moved in the opposite direction. On Friday, October 30, Google, Microsoft and OpenAI announced that they would honor one shared rule in robots.txt concerning AI training. Instead of separate entries for each company’s bot, a group named __usage_ai-training with a Disallow rule would be enough. The idea comes from a project that Kevin Kelley of Anthropic submitted to the IETF working group, the organization that develops internet standards. It is an individual draft, not an official standard.
None of the companies support it yet, and their timelines differ. Google plans support at the beginning of 2027 and, if a website does not have a separate rule for Google-Extended, it will treat the shared entry as a block for Google-Extended, which will also exclude the content from grounding, that is, providing it to the model when generating responses in Gemini applications. Microsoft separates training from grounding, will start honoring the new setting on February 1, 2027, and after March 1, 2027, will stop recognizing the old NoArchive and NoCache tags. OpenAI says within the next few months. Meta is not on that list.
What this means for companies
Our conclusion: robots.txt is becoming the basic tool for managing what AI is allowed to do with a company’s content, but it will not replace a contract. The dispute with Meta shows that consent written into a platform’s terms may go further than a technical block on the website. The ruling in Oslo and the proceedings in London suggest that regulators are treating this problem seriously, but a Polish company today has no guarantee that UOKiK will act before November 20.
Before that date, it is worth doing a few things:
The decision whether to stay on Meta’s platforms depends on how much the company owes to those channels and how much it loses by giving away content without compensation. Everyone has to do that calculation for themselves, but it is better to do it before November 20 than after.