Meta has paused, indefinitely, a rewrite of the business terms that millions of advertisers, shops and publishers accept simply by using Facebook and Instagram. The draft would have let Meta use content from its business customers' websites to build and improve its products, made customers "solely responsible" for what Meta's AI features produced in their name, and treated continued use as acceptance. After a court order in Norway, a UK competition investigation and a loud backlash from publishers and advertisers, Meta's Commercial Terms page now says the updates "are on hold" and the existing terms "remain in effect until further notice".
The climbdown is welcome. The question it leaves behind is bigger than Meta: when does AI innovation become a way of shifting legal risk onto customers?
In brief
Key takeaways
- What Meta proposed: updated Commercial Terms, Product Catalog Terms, Self-Serve Ad Terms and Ad Creative Generative AI Terms that, according to legal analyses of the draft, would have let Meta use content from business accounts' domains, web pages and URLs for its products, and placed responsibility for AI-generated content on customers.
- What happened: the changes, first due on 30 October 2026, were pushed to 20 November, softened, and then paused indefinitely in early November. Meta says it is pausing them "as we continue to engage with stakeholders and take our time to get this right."
- Who forced it: ten Norwegian media groups won an interim injunction in Oslo on 28 October; on 30 October the UK's Competition and Markets Authority opened a Competition Act investigation into whether Meta is imposing exploitative terms on publishers and advertisers. Meta says there are "no grounds" for that investigation.
- What hasn't changed: the existing terms, dated March 2024, already require businesses to indemnify Meta against third-party claims. The pause stops the expansion; it does not reset the balance.
- The wider lesson: when a platform's AI writes your ads and reads your website, the contract decides who pays when it goes wrong. That is not a technical detail. It is the business model.
1. What did Meta's new AI terms actually say?
Terms of service rarely make news. These did because of three changes that business customers and their lawyers spotted in the draft.
- Your website as raw material. As drafted, businesses would have permitted Meta to use content from their domains, web pages or URLs to "provide, develop, improve, research, train and maintain the integrity" of Meta products, according to reporting on the draft text and legal alerts from firms advising advertisers. Commentators read that as permission to crawl and reuse sites linked to a business account, with no way to opt out.
- AI mistakes become your mistakes. A section on AI features reportedly warned that they "may make mistakes" while making the customer solely responsible for the resulting content. For advertisers using Meta's automated tools, which can rewrite copy, generate images and add features to existing ads, that meant carrying the legal risk for output they did not write and, in some cases, features switched on by default.
- Acceptance by default. As with most platform terms, continuing to use Meta's business tools would count as acceptance. There was no way to accept part and reject the rest. The only opt-out legal commentators identified was a 30-day window for US businesses to opt out of arbitration.
Before pausing, Meta removed the word "train" from the list of permitted uses and narrowed the scope from all Meta products to its advertising and business products. Critics noted that "develop, improve, research and maintain" still covered a great deal.
2. How did the backlash force Meta to back down?
Meta did not reverse course because of angry posts. It reversed course because the backlash found legal teeth in two places at once.
Norway. Ten Norwegian media groups, including the public broadcaster NRK and Schibsted, asked the Oslo District Court to block the terms, arguing they would let Meta take journalism to feed its AI without paying for it. On 28 October the court granted interim relief, barring Meta from applying the new terms to Norwegian Facebook and Instagram business accounts until their legality is decided. The order covers Norway only.
The UK. Two days later the Competition and Markets Authority opened an investigation under the Competition Act 1998 into suspected abuse of a dominant position, naming Meta Platforms, Inc., Facebook UK Limited and Facebook Ireland Limited. The question is whether Meta is imposing exploitative terms on publishers and advertisers, including terms that could let it use content from publishers' websites to build its own products and AI features. The CMA acted on concerns raised by groups including the News Media Association and Movement for an Open Web. It stresses that no decision has been made on whether the law has been broken, and it expects to gather information until January 2027.
Facing an injunction in one country and an antitrust case in another, Meta first delayed the start date to 20 November, then paused the changes altogether. Its statement was short: "We're pausing the changes as we continue to engage with stakeholders and take our time to get this right. In the meantime, our existing terms remain in effect."
- sets of Meta business terms whose updates are on hold
- 4
- Meta; press reports
- Norwegian media groups that won an interim injunction in Oslo
- 10
- Oslo District Court, via press reports
- Meta companies named in the UK CMA's Competition Act investigation
- 3
- UK Competition and Markets Authority
- the only opt-out window identified in the draft, for US arbitration
- 30 days
- Legal analyses of the draft terms
3. Why is shifting AI liability onto customers so controversial?
Every contract allocates risk. That is its job. What made this one controversial is the direction the risk was travelling and who controls it.
Consider how an advertiser uses Meta today. Increasingly, Meta's systems choose the audience, write variations of the headline, generate images, crop video and decide where the ad appears. The more of that work the platform's AI does, the less of the final ad the advertiser actually authored. Yet the draft terms, as reported, would have made the advertiser solely responsible for the result.
That is the inversion at the heart of the story. The party with the most control over the AI would carry the least risk from it, and the party with the least control would carry the most.
Defenders of Meta make three fair points. First, businesses are not forced to use AI features. Second, advertisers have always been responsible for what their ads claim; that principle did not start with AI. Third, Meta's existing terms already require businesses to indemnify it against third-party claims, so this was partly a clarification rather than an invention.
Each point has an answer. AI features are becoming the default rather than an option, and some reportedly arrive switched on. Responsibility for claims you make is different from responsibility for words a machine wrote for you. And extending an old indemnity to cover a new, unpredictable technology is not a small clarification; it is a transfer of an entirely new category of risk.
4. When does AI innovation become risk-shifting?
Here is my test. AI innovation becomes risk-shifting when three things happen together.
- The vendor controls the system. The customer cannot inspect the model, audit its training data or predict exactly what it will produce.
- The vendor disclaims the output. The terms warn that the AI "may make mistakes" and decline responsibility for them.
- The customer cannot realistically say no. Acceptance is automatic, the terms cannot be negotiated, and leaving the platform means losing access to the customers who live there.
When all three are true, "innovation" is doing something more than improving a product. It is moving the cost of the product's failures off the vendor's balance sheet and onto the customer's. The technology is new; the manoeuvre is old. Industries with market power have always written contracts that keep the upside and outsource the downside.
What makes AI different is scale and opacity. A misleading claim in a single ad is a contained problem. An AI system rewriting millions of ads, or summarising millions of web pages, can make the same mistake millions of times, and the customer may never see the version the system actually served. Liability that follows authorship made sense when humans wrote the ads. It makes much less sense when the platform's software does.
If you sell the intelligence, you should not get to disown the mistakes.
5. Why are publishers so angry about the website clause?
For news organisations, the content clause mattered even more than the liability one. Journalism is expensive to produce, and AI companies have spent the past few years either signing licensing deals with publishers or fighting them in court over whether their content can be used without payment.
A term that let Meta use material from any website linked to a business account, simply because the publisher promotes its stories on Facebook, would have side-stepped that negotiation entirely. Publishers would have faced a choice between giving up control of their content and giving up one of their largest sources of readers. That is why the Norwegian groups went to court and why the CMA framed its case around exploitative terms imposed by a dominant platform.
There is a principle here that applies well beyond media. Consent obtained because the alternative is commercial exclusion is not much of a consent. A platform that is also a gatekeeper has a special duty not to turn access into leverage over everything else its customers own.
6. What should businesses do now?
The pause is a reprieve, not a resolution. Meta has said it will keep engaging with stakeholders, and similar terms may return in a different form. Businesses that rely on Meta, or on any AI platform, can use the time well:
- Read the AI sections of every platform contract you accept. Look for who owns the inputs, what the vendor may do with your content, and who is liable for generated output.
- Audit default-on AI features. If a platform can rewrite your ads automatically, check which tools are enabled in your account and turn off what you have not approved.
- Keep records of what you actually authored. If liability ever turns on who wrote what, the business that can show its approved version will be in a far stronger position.
- Treat your website as an asset. Decide deliberately which crawlers and partners may use your content, and say so in your own terms and technical settings.
- Use your collective voice. This climbdown happened because publishers, advertisers and trade bodies acted together and regulators listened.
7. The bigger picture
The AI boom has produced plenty of arguments about who controls the technology and who profits from it, from the chips at the bottom of the stack to the fight over open and closed AI models. Contracts like Meta's are where that argument becomes concrete for ordinary businesses. They decide whose content feeds the machines and who pays when the machines get it wrong.
Meta backed down this time because a court and a regulator gave its customers leverage they did not have on their own. The lesson for every platform building AI into products that other businesses depend on is simple: if the only way to make your innovation work is to make your customers carry its risk, the innovation is not finished.
Questions readers ask
Frequently asked questions
What were Meta's new AI terms? They were planned updates to Meta's Commercial Terms, Product Catalog Terms, Self-Serve Ad Terms and Ad Creative Generative AI Terms. According to legal analyses, they would have let Meta use content from business customers' websites for its products and made customers solely responsible for content produced by Meta's AI features.
Has Meta cancelled the new commercial terms? Meta has paused them indefinitely rather than cancelled them. Its Commercial Terms page says the updates are on hold and the existing terms remain in effect until further notice. Meta says it is continuing to engage with stakeholders.
Why did a Norwegian court block Meta's terms? Ten Norwegian media groups argued the terms would let Meta use their journalism to build AI without payment. On 28 October 2026 the Oslo District Court granted interim relief barring Meta from applying the new terms to Norwegian business accounts while their legality is decided.
Is the UK investigating Meta over its AI terms? Yes. On 30 October 2026 the Competition and Markets Authority opened a Competition Act investigation into whether Meta is abusing a dominant position by imposing exploitative terms on publishers and advertisers. The CMA says no decision has been made on whether the law has been broken.
Are advertisers liable for mistakes made by Meta's AI? Under the paused draft, advertisers would reportedly have been solely responsible for content produced by Meta's AI features. Under the existing terms, businesses already indemnify Meta against third-party claims, so advertisers should check which AI features are enabled and what they approve.
Can businesses stop Meta using their website content? The draft terms offered no opt-out for content use; the only opt-out identified was a 30-day window to opt out of arbitration for US businesses. With the update paused, the existing terms apply.
Sources
Sources
- Meta: Commercial Terms (status notice: updates on hold)
- Meta: Product Catalog Terms
- UK Competition and Markets Authority: CMA opens investigation into Meta's commercial terms (30 October 2026)
- UK Competition and Markets Authority: case page, suspected abuse of a dominant position in relation to Meta's commercial terms
- Computerworld, "Meta does a U-turn on controversial new AI terms for businesses", November 2026
- Nieman Lab, "Meta's new commercial terms will allow the scraping of news sites, legal experts say", October 2026
- Shumaker, Loop & Kendrick LLP, client alert, "Meta's New Commercial Terms: Meta Gets All the Data, Enterprise Customers Get All the Liability", October 2026
- European Broadcasting Union, statement calling on Meta to revise its Commercial Terms and provide a genuine opt-out, October 2026
- Reporting on the Oslo District Court interim injunction sought by ten Norwegian media groups, 28 October 2026
Editor's note: This is an opinion piece; views are the author's own. Descriptions of the draft terms are based on legal analyses and press reporting of the text Meta previewed; the changes have not taken effect. The UK CMA investigation is at an early stage and has made no finding, and Meta says there are no grounds for it.





