Meta Says Its AI Ad Edits Are Your Problem. One Court Disagrees

Meta's updated advertising terms take effect October 30, 2026, and they make advertisers "solely responsible" for whatever Meta's AI features create or change in their ads, including features Meta switches on by default. Meta's own liability stays capped at the greater of $100 or 12 months of fees. Every auto-enabled Advantage+ enhancement needs a review before Black Friday creative goes live.

The old arrangement was at least logically tidy. If you chose to use Meta's AI creative tools, the output was your problem. According to Adweek's report on the new terms, that condition is gone. A new "AI Features" section puts the responsibility on the advertiser regardless of who flipped the switch, and Meta concedes in the same text that the features "may make mistakes and may not always produce accurate, complete, or suitable results."

So Meta is telling you, in writing, that the tool it turns on for you will sometimes get things wrong, and that when it does, you're holding the bag. I've read a lot of platform terms. This is one of the more honest ones, honestly, which is part of what makes it uncomfortable.

Three clauses that matter more together than apart

PPC Land's breakdown lines the new AI language up against the terms that are already in force, and the combination is the real story. Any one of these clauses on its own is standard boilerplate. Stacked, they read differently.

• Acceptance is passive. Section 6 of the current terms says that "continuing any access or use" of Meta's business products counts as agreeing to updates. No click, no signature. If an ad is running on October 30, you've accepted.
• Meta's exposure is capped. Section 4 limits Meta's liability to the greater of $100 or what you paid Meta in the past twelve months.
• Your exposure isn't. Section 5.a has advertisers indemnify Meta against third-party claims related to their content on Meta products, legal fees included.

Put those next to the new AI section and the shape is pretty clear. Meta generates or alters the creative, the advertiser owns the output, and if someone sues over it, the indemnity clause points the bill back at the advertiser. PPC Land also notes Section 1 extends the licence to content placed by "your agency that places an ad for you," which drags agencies into the same chain.

Meta's line is that this is housekeeping, folding its existing generative AI ad terms into the main document, and that advertisers can turn off individual Advantage+ features in Ads Manager. What Meta didn't answer, per Adweek, is the obvious question: are you liable for a feature you never enabled? The text seems to say yes. Meta declined to say directly.

Why the timing looks less like housekeeping

Here's the part I keep coming back to. In March, a federal judge in California ruled in Bouck v. Meta Platforms that Section 230 might not protect Meta from claims over scam ads its generative AI tools helped create. Katten's analysis of the decision walks through the court's reasoning: if a scammer asks Meta's tool for an investment ad and the tool writes new fraudulent wording on its own, that language is "all Meta," and Meta starts to look like the creator of the content rather than a host.

It's a pleading-stage ruling. It isn't binding on other courts, there are no factual findings, and it could be reversed. Katten is careful about all of that, and I should be too (I'm not a lawyer, and this isn't legal advice).

Still. A court floats the idea that a platform could be directly liable for what its AI writes, and seven months later the platform rewrites its contract so the advertiser is responsible for what its AI writes. Maybe that's coincidence. If I were in Meta's legal department, though, I'd want exactly this clause.

The AI is Meta's. The signature on the ad is yours.

This already went wrong, in public, for brands that didn't opt in

None of this is hypothetical. In June, REI pulled an Instagram ad for a Van Rysel bicycle after people spotted an extra set of handlebars, a missing crank arm and brakes on the wrong side. PPC Land's reconstruction found the ad ran for roughly a week, and REI told Fast Company that "Meta auto-enrolled us in an AI personalization tool" without its knowledge.

The Adweek piece adds more. Bookshop.org said on Threads that an account update opted it into a generative feature that added an AI illustration to a human-made ad, and it has since turned off every Meta AI ad feature. Ryan Adams at Exverus described Meta adding sitelinks, translating copy and building AI images and carousels from brand media, sometimes without the advertiser enabling it, and said the agency has pulled live ads to strip the changes. Kayla Loecker at AdCellerant said they now turn off the default AI automation on every campaign.

Under the old terms, REI's bike was arguably Meta's mess. From October 30, it's REI's, contractually at least. And the reputational damage was always going to be the brand's either way. Asa Hiken at Emarketer made the point that consumers don't know Meta's ad policies. They see a broken bike with your logo on it.

The settings don't stay off, which is the real problem

"Just turn the features off" sounds simple. It mostly isn't. A JumpFly write-up from August found that replacing the creative, changing the ad format or duplicating an ad can cause recommended enhancements to show as selected again. There's no single switch that kills all Advantage+ automation across a campaign. Adams told Adweek many options are "tucked away in hidden or collapsed menus," which matches what most media buyers I've seen complain about.

So it's a bit like signing for a rental car where the company can swap the tires overnight, and any blowout is on your insurance. You can check the tires every morning. You just have to actually do it, every morning, during the busiest five weeks of the year.

And to be fair to the features, some of them work. Jon Loomer's rundown of the enhancements leaves most of them on by default and only routinely switches off Image Template, while flagging things like Relevant Comments as worth disabling for sensitive clients. Plenty of these edits are fine. What changes on October 30 is that you're accountable for edits you may never have seen, and that shifts the math on the ones that only add a sliver of performance. (PPC Land notes one incrementality test credited Advantage+ with just 17% of the conversions Meta reported, for whatever that's worth.)

A pre-October 30 audit that fits in an afternoon

Here's what I'd do before the 30th, roughly in this order:

1. Set the account-level default first. JumpFly points to Ad account settings, then Creative features, as the place to limit automatic opt-ins to new creative features. Do this before touching individual ads, or you'll be fixing ads while new ones get enrolled.
2. Open the full Advantage+ creative menu on every active ad. Not just the top toggle. Expand every collapsed section. Your benchmark: zero active ads with an enhancement you can't name and haven't approved. If you run more than 50 ads, start with the top 20 by spend, which is usually most of the risk.
3. Kill anything that generates new imagery or text for regulated categories. Finance, health, supplements, anything with required disclaimers or price claims. A hallucinated price or an "enhanced" before/after shot is exactly the kind of thing a third party sues over.
4. Re-check after every duplicate or creative swap. Put it in the Q4 launch checklist so nobody skips it at 11pm on Cyber Monday.
5. Screenshot your settings with a date stamp. If something goes wrong later, you want evidence of what you had switched off and when.
6. Agencies: read your client contracts this week. If Meta's indemnity lands on the advertiser and your agreement says you're responsible for creative, you may be the one actually holding it.

Go Fish Digital's David Dweck told Adweek they won't advise clients to pull back in Q4 and called the new terms "the tax of advertising on Meta." I think that's probably right for most accounts. Leaving Meta in November isn't a real option for a lot of brands. Auditing for an afternoon is.

This also isn't just a Meta thing. ChatGPT Ads is turning on its own creative rewrites by default around the same date, which we covered in ChatGPT Ads Starts Rewriting Your Ads Oct 30. Meta's same terms update also covers website crawling, which we unpacked in Meta's New Terms Let It Crawl Your Site From Oct 30, and product browsing gets switched on in existing ads two days earlier (our breakdown of that default).

Where I think this goes after the holidays

I keep wondering whether the full text, once it's public, will be softer than the coverage suggests. PPC Land couldn't retrieve Meta's preview to verify the new section, so for now we're working from Adweek's quotes. That's a real gap, and I'd rather say so than pretend otherwise.

My guess: somewhere in the first half of 2027, a brand ends up in a public dispute over an AI-altered Meta ad it says it never approved, and that pressure gets Meta to ship a version-level log showing exactly which AI variants ran and when. I'd put that at maybe 60%. Until then, the only record of what your ads actually looked like is the one you keep yourself, so it's probably worth starting it before the 30th rather than after the first screenshot of a three-handled bike shows up with your name on it.

By Notice Me Senpai Editorial