Meta Indefinitely Pauses AI-Training Terms After Publisher Backlash and Court Injunctions in Norway and Brazil

Meta has indefinitely paused a controversial rewrite of its commercial terms that would have allowed the company to send automated software to the websites of businesses using its advertising tools and to use what it collected, including to train artificial intelligence, after three weeks of mounting opposition from news publishers, politicians, regulators and courts in three countries.

On Thursday 5 November, the company removed the preview link to the proposed terms and posted a notice in its Business Help Center saying the updates to four sets of terms — the Meta Commercial Terms, Product Catalog Terms, Self-Serve Ad Terms and Ad Creative Generative AI Terms — “are now paused and will not take effect on November 20, 2026”. Meta confirmed the indefinite delay in a statement to Nieman Lab, telling the publication: “We’re pausing the changes as we continue to engage with stakeholders and take our time to get this right.”

The pause leaves the existing terms in force. Meta gave no new date, said nothing about which provisions it intends to rewrite, and described the step as a pause rather than a withdrawal — leaving open the possibility that the same text, or a modified version, returns at a later date.

For news publishers, who led the opposition, the retreat is a significant victory. The draft terms would have reached every company that advertises with Meta, but publishers argued that their archives — the journalism they spend money to produce — were the content most exposed to being harvested and reused without asking, and without payment.

What the terms would have allowed

The proposed terms required every business using Meta’s products to authorise web-scraping tools, and test credentials, on the domains associated with its products or services. Continued use of Meta’s business tools would have counted as acceptance. According to a reading of the text published by PPC Land on 17 October, the new terms granted Meta the right to crawl advertisers’ websites for artificial intelligence.

That scope was what made the dispute so large. The contract made no distinction by sector. It reached the website of any business with a Facebook page — retailers with product catalogues, app developers, agencies running campaigns for clients, and news publishers alike. Publishers led the opposition because their archives are the material most valuable to an AI system hungry for text, but the clause itself applied to everyone.

Meta owns Facebook, Messenger, Instagram, WhatsApp and Threads, according to Nieman Lab, and its business tools sit at the centre of how millions of companies reach customers. For most of those companies, walking away from Meta’s advertising platform is not a realistic option, which is why the terms were described by critics as a take-it-or-leave-it bargain: accept the crawling, or stop using the tools.

The package contained a second provision that drew less attention but mattered directly to advertisers. According to Adweek, reported on 22 October, the revised Commercial Terms included a new section on AI features that would have made advertisers solely responsible for Meta’s AI edits to their ads, including features Meta switches on by itself. One agency told Adweek it had pulled live advertisements to strip out sitelinks, translations and AI-generated images added by Meta’s Advantage+ features. With all four updates now on hold, that consolidation, and the liability language attached to it, will not take effect on 20 November either.

The scale of the population covered is large. Meta’s chief financial officer, Susan Li, said on the company’s third-quarter earnings call on 21 October that nearly 10 million advertisers used at least one generative AI creative tool during the quarter.

Meta has maintained throughout that its crawling practices remain unchanged and that content owners can still use robots.txt — the file websites use to signal which automated visitors they do and do not want — to state their preferences. Critics reply that the revised terms never explicitly committed Meta to respect those signals, that a robots.txt file is no substitute for a contractual right to refuse, and that the purposes for which collected material could be used were drawn so broadly that they covered almost anything Meta might later decide to do with it.

Three weeks from first report to pause

The dispute moved quickly. Nieman Lab first reported on the terms on 15 October. They were initially set to take effect on 30 October, “sparking global outcry among news publishers, politicians, and trade organizations”, according to the publication.

On 16 October, Meta removed the word “train” from the terms and limited the use of content scraped in future to “advertising and business products”. Nieman Lab reported on 17 October that legal experts considered the revised wording to still grant broad scraping permissions. The record on the exact timing is not entirely consistent — when PPC Land checked the UK English preview on 18 October, the AI training sentence was still in place, and the São Paulo court that later ruled on the terms dated the deletion to 19 October — but the substance of the change was the same: the explicit training language went, and the permission to collect stayed.

On 21 October, Nieman Lab reported that politicians in the United Kingdom and publishers in Germany had called on regulators to pause the terms before the original 30 October deadline. On 22 October, the News/Media Alliance, the trade body for American news publishers, gave Meta until 26 October to answer publicly and demanded that the terms be withdrawn. The same day, Adweek reported the AI-features liability section described above.

On 23 October, Meta moved the effective date back by three weeks, to 20 November. On 24 October, the European Publishers Council asked the European Commission to halt the crawling terms before 20 November and, failing interim measures, to obtain an immediate undertaking from Meta to postpone the contested provisions.

On 27 October, Nieman Lab reported that Senator Adam Schiff, a California Democrat, had written to Meta asking it to scrap the terms, describing them in his letter as “a gross abuse of Meta’s power”. On 28 October, Andy Stone, Meta’s vice president of communications, posted on X: “We’ve heard publishers’ feedback on our proposed ads and business terms update. We care deeply about our partners and do not plan to make policy changes without a full consultation with industry groups and leaders. We’re committed to getting this right, and as we expect this consultation could extend beyond November 20, in which case we’d delay implementation of the updated terms.”

The same day, the dispute moved from letters to formal proceedings. The Oslo District Court temporarily blocked Meta’s new terms after ten media groups sued through the Norwegian Media Businesses’ Association, a group that includes NRK, TV 2, Schibsted and Amedia.

On 30 October, the United Kingdom’s Competition and Markets Authority opened a formal investigation into Meta, examining whether the company was “abusing a dominant position by imposing exploitative terms on publishers and advertisers”. That was also the original effective date of the terms. On 31 October, Google and Bing said they would back a single robots.txt line letting website owners opt out of AI training — a shared consent signal of exactly the kind Meta’s terms had not offered.

At the end of October, Senators Adam Schiff and Marsha Blackburn, a Tennessee Republican, each sent letters to Meta chief executive Mark Zuckerberg asking him to withdraw the terms. On 3 November, a state court judge in São Paulo suspended the crawling clause for members of Brazil’s three national media associations and attached a penalty under which Meta faces a fine of one million reais a day if it scrapes Brazilian media sites for AI, capped at 100 million reais.

Two days later, on 5 November, Meta took down the preview link and posted the pause notice — 21 days after Nieman Lab first reported the terms.

Two courts, one regulator and two senators

The two court orders are the sharpest part of the story. In Oslo, the order came on 28 October, when the district court temporarily blocked the new terms after ten media groups sued through the Norwegian Media Businesses’ Association. In São Paulo, a state court judge on 3 November suspended the crawling clause for members of Brazil’s three national media associations — ANJ, Abert and Aner — and attached the daily fine of one million reais, capped at 100 million reais, if Meta scrapes Brazilian media sites for AI. Both orders were granted without Meta being heard, and both were framed to hold until the courts decide the cases, not until Meta finishes consulting.

The United Kingdom’s Competition and Markets Authority opened its formal investigation on 30 October, examining whether Meta was “abusing a dominant position by imposing exploitative terms on publishers and advertisers”, according to Nieman Lab. The CMA’s case page covers Meta’s existing commercial terms as well as the revisions, with an initial information-gathering phase scheduled to run until January 2027. A pause of the revised text therefore does not, on the face of the case page, remove the subject of the investigation. The regulator is looking at the structure of the bargain itself, not only at the new wording.

In the United States, the terms “stoked bipartisan concern”, according to Nieman Lab. Senators Schiff and Blackburn, from opposite parties, both wrote to Zuckerberg at the end of October asking him to withdraw the terms. Theo Bamber, chief executive of the United Kingdom’s News Media Association, asked on LinkedIn: “Does Meta think we can’t read?” and called the company’s “continued arrogance” “staggering”.

Trade bodies had pressed the same point from the start. The News/Media Alliance gave Meta until 26 October to answer publicly and demanded withdrawal. The European Publishers Council asked the European Commission to halt the crawling terms before 20 November. On 21 October, Nieman Lab reported that politicians in the United Kingdom and publishers in Germany had called on regulators to pause the terms before the original 30 October deadline.

The 5 November pause meets the timing those groups asked for. On the evidence available, it does not meet the form. It is a notice Meta can revise on its own, not a commitment given to a court or a regulator, and the Nieman Lab report does not say whether Meta will ask the Oslo or São Paulo courts to lift their orders, whether the CMA will continue as planned, or whether either senator has received a reply.

What stays in force

“In the meantime, our existing terms remain in effect,” Meta’s notice says. For the Commercial Terms, that means the version that has applied since 6 March 2024. The existing Self-Serve Ad Terms, Product Catalog Terms and generative AI ad terms also continue to apply.

The practical consequence for businesses is that 20 November is no longer a date on which continued use of Meta’s tools would amount to accepting new text. The notice asks nothing of advertisers, publishers or merchants. It says nothing about Meta’s crawlers, such as Meta-ExternalAgent, or about how the company treats robots.txt instructions. The pause concerns the contract, not the software.

Nor does the pause settle the substance. The revised clause, in whatever form it last took, has not been withdrawn. The objections publishers raised — that a robots.txt file is no substitute for a contractual right to refuse, that the purposes of collection were drawn broadly, and that the only way to reject the terms was to stop using Meta’s business products — have not, on the record so far, received a published answer from Meta.

What the notice leaves out is as telling as what it says. It does not mention website crawling, publishers, artificial intelligence training or any specific clause. It does not name the stakeholders Meta is consulting, describe how the consultation works, or set a deadline for it. Meta’s statement to Nieman Lab adds only that the company wants to “take our time to get this right”.

Analysis: Why It Matters

The pause is worth more attention than a routine corporate climbdown, because of what it reveals about where the power in the AI economy now sits — and about who is finally able to push back against it.

A rare defeat for the platform playbook

For a decade, the standard move by a dominant platform has been to change the contract and dare the other side to leave. The terms are presented as a package, acceptance is counted by continued use, and the cost of refusing is exclusion from the marketplace where the customers are. Meta’s draft followed that playbook exactly: every business using its products would authorise crawling, and the only way to say no was to stop using the tools.

What is new is that the playbook failed, and failed quickly. Twenty-one days passed between the first report and the pause. In that time, publishers organised across borders, two courts intervened, a competition authority opened a formal case, and senators from both American parties wrote to the chief executive. The speed matters. Platform terms usually change slowly and are challenged slowly; here, the challenge arrived before the terms had even taken effect. That is a change in the balance of power, and other platforms will have noticed it.

The fight is about consent, not just copyright

Much of the public debate about AI and journalism has been framed as a copyright argument: who owns the words, and who should be paid for them. The Meta dispute is different, and in some ways more important. It is about consent — about whether a company that controls the marketplace where a business must operate can also write itself permission to take that business’s work as a condition of staying in the marketplace.

That distinction is why the opposition was so broad. Retailers, app developers and agencies were covered by the same clause, even though publishers led the campaign. The objection was not only that journalism might be used to train AI systems. It was that a contract could convert a business relationship — “you advertise with us” — into a data pipeline — “we may collect everything on your website” — without the business ever signing anything new. Continued use would have been the signature.

There is a constructive alternative, and it is already being demonstrated elsewhere. Anthropic, the maker of the Claude AI models, began paying for news for the first time when it licensed election data from the Associated Press for Claude — a deal reported earlier this month. Licensing treats journalism as something to buy, not something to collect. Meta’s draft treated it as something to collect. The difference between those two approaches is the difference between a market and a taking, and it is the central question in the AI economy.

Courts moved faster than regulators — and faster than Meta expected

The two injunctions deserve attention in their own right. Courts in Norway and Brazil acted within days of the lawsuits being filed, and both acted before the terms took effect. That is unusual. Courts are normally the slowest instrument available to a challenger, and companies usually count on that slowness: by the time a case is heard, the new reality is already in place and hard to undo.

Here the sequence ran the other way. The Oslo order came on 28 October, the same day Meta’s own communications chief was promising a full consultation. The São Paulo order came on 3 November, two days before the pause. Both orders were granted without Meta being heard, which is a measure of how seriously the judges took the risk of irreversible harm: once a website has been crawled and its contents absorbed into a training run, the material cannot be un-collected.

The lesson for other companies is plain. A clause that looks like a routine commercial term can be treated by a court as an urgent matter if the harm it enables cannot be undone. The AI industry’s habit of moving first and litigating later has just met a counter-example.

The robots.txt gap is the real story

Behind the headlines about courts and senators sits a technical question that will decide much of the next decade: what does a website owner’s “no” actually mean, and who has to respect it?

Meta says content owners can use robots.txt to signal their preferences. Critics say the revised terms never explicitly committed Meta to respect those signals. That gap is the whole argument in miniature. A robots.txt file is a voluntary convention — a polite request that a well-behaved crawler may honour. A contract term is a binding right. Meta’s draft would have given itself a contractual right to collect, while offering publishers only a voluntary signal in return. The asymmetry was the point of the objection.

The industry is now moving, unevenly, towards a shared answer. On 31 October, Google and Bing said they would back a single robots.txt line letting website owners opt out of AI training. That is a step towards making “no” mean something in practice. But a shared signal only works if the biggest collectors honour it, and Meta’s pause says nothing about its crawlers or about how it treats robots.txt instructions. Until Meta answers that question in writing, publishers have won a delay, not a settlement.

What it means for publishers — and for every business on Facebook

For news publishers, the pause buys time and leverage. The terms will not take effect on 20 November, the court orders remain in place, and the CMA investigation continues. Publishers now negotiate from a position in which Meta has already conceded, in public, that it needs to “get this right”. That is a stronger position than the one they held on 15 October, when the terms were simply announced.

For every other business on Facebook, the implications are just as real. The crawling clause was never a publishers-only provision. A retailer whose product catalogue was scraped, or an agency whose clients’ websites were crawled, would have had the same complaint and less organised support. The pause protects them too, for now.

But the protection is temporary. The notice is Meta’s own, it can be revised by Meta alone, and it contains no commitment about what the replacement terms will say. Publishers and businesses should treat the pause as a window in which to demand a written answer to the questions Meta has not yet answered: will the company commit to respecting robots.txt signals? Will it exclude AI training from any future crawling permission unless a website owner agrees? Will it pay for the material it collects, as Anthropic has begun to do with the Associated Press?

The wider AI backlash is now a political fact

The Meta dispute does not stand alone. It arrives in the same weeks that American voters rejected data-centre expansion in 40 of 41 local ballot measures during the midterm elections, according to CNBC, and that investors cut their expectations for Anthropic’s planned stock market debut from a two-trillion-dollar valuation towards one to one-and-a-half trillion dollars, according to CNBC, Bloomberg and the Wall Street Journal. The AI industry is discovering that public consent cannot be assumed — not for the buildings that house its computers, not for the valuations it asks investors to believe, and not for the data it wants to collect.

That is a healthy development. AI systems are built on three things: computing power, energy and data. All three are now contested in public, by voters, by courts and by regulators. The industry’s answer to that contest will shape whether AI becomes a technology that societies choose, or one that is imposed on them. Meta’s pause is a small but real sign that the choosing has begun.

What to Watch Next

1. Meta’s “further update”. The company has promised to provide one “when we have more to share”. The wording of the replacement terms — whether they drop the crawling permission, limit it, or return it in modified form — will show whether the pause was a retreat or a delay.
2. The Oslo and São Paulo cases. Both injunctions hold until the courts decide the cases. Watch whether Meta asks either court to lift its order, and whether the Brazilian fine of one million reais a day is ever tested.
3. The CMA investigation. The United Kingdom’s Competition and Markets Authority is gathering information until January 2027, and its case covers Meta’s existing terms as well as the revisions. A finding that Meta abused a dominant position would have consequences far beyond this one dispute.
4. Washington. Senators Schiff and Blackburn wrote to Mark Zuckerberg asking him to withdraw the terms. Whether either receives a reply, and whether Congress moves from letters to legislation, will show whether the bipartisan concern hardens into law.
5. The robots.txt standard. Google and Bing have backed a single line letting site owners opt out of AI training. Whether Meta adopts the same signal — and commits to honouring it in writing — is the clearest test of whether the pause means anything in practice.
6. Licensing versus scraping. Anthropic’s deal to license Associated Press election data for Claude shows that paying for news is possible. Watch whether Meta, or any other platform, follows that model instead of returning to a contract that takes the material for free.

Sources

• Nieman Lab — Meta pulls down controversial terms of service (5 November 2026)
• Reuters, via TradersUnion — Meta pauses commercial terms update after AI training backlash (6 November 2026)
• PPC Land — Meta halts its Nov 20 terms that would let it crawl business websites
• TokenPost — Meta Pauses Terms Update That Would Enable AI Use of Facebook-Linked Content (8 November 2026)
• AI Weekly — Meta Pauses Nov 20 Scraping Terms After Norway, Brazil Rulings
• ad-hoc-news — Meta Platforms stock pauses AI terms after publisher backlash (6 November 2026)