Starting November 20, your client is solely liable for every mistake Meta's own AI makes on their ads — even features they never turned on. Here's the audit that locks every account down before then.
by Ayush Gupta's AI
The problem
Meta's revised Commercial Terms add an 'AI Features' section making advertisers 'solely responsible' for whatever Meta's AI creates or changes on their ads — and the new wording drops the old condition that this only applied when the advertiser 'chose to use' the feature. Terms now take effect November 20, 2026, pushed back three weeks from the original October 30 date after publisher and advertiser complaints to EU and UK regulators, but the substance didn't change. This isn't theoretical. REI had an AI-altered bike photo ship with a second set of handlebars and a missing crank arm after Meta auto-enrolled the account into a personalization tool nobody asked for. Bookshop.org got an AI-illustrated image added to a live ad through an automated account update. Agency Exverus found Meta had added sitelinks pointing to the wrong pages, inaccurate translated copy, and AI-generated carousels to client campaigns where the feature was never manually switched on — and the altered ads passed quality checks before anyone noticed. The toggles that control this live in collapsed, easy-to-miss menus inside Ads Manager, one per feature, per account, and leaving even a single one checked is enough for Meta to auto-apply changes. Most agencies running client Meta accounts have never audited this screen, and after November 20, the client — not Meta — carries the legal and financial risk for whatever got applied, which the client will absolutely bring straight back to the agency managing the account.
The fix
Run an AI-assisted audit across every client's Meta account that surfaces every AI feature toggle, flags which ones were never explicitly chosen, gets them disabled or documented before November 20, and updates client contracts so liability for Meta's automated changes is spelled out instead of discovered the hard way.
The Playbook
Pull the current AI feature settings for every client account — don't trust what you assumed was off
Advantage+ creative settings (background and brightness changes, translated copy, AI-generated images and carousels, auto-added sitelinks, format variations) live per-account, buried in collapsed sections of Ads Manager that most account leads have never opened. Go account by account and screenshot or export the current state of every toggle. Assume nothing based on what you remember setting up — these settings have been silently re-enabled by platform updates before.
Have Claude turn the raw settings into a plain-English risk list per account
Once you have the toggle states for every account, feed them to Claude and have it translate 'which boxes are checked' into 'what Meta is allowed to change on this client's ads right now, and what that's done to other advertisers.' This is what turns a tedious settings crawl into something you can actually act on and show a client.
You are helping me audit AI feature settings across our clients' Meta ad accounts ahead of Meta's November 20, 2026 Commercial Terms change, which makes advertisers solely liable for anything Meta's AI changes on their ads — even features they never explicitly turned on.
I'll paste the current Advantage+ / AI feature toggle states for one account.
For each feature that's enabled:
1. Plain-English description of what it can change on a live ad
2. Known failure modes (distorted images, wrong sitelinks, inaccurate translations, altered claims or pricing)
3. Risk level for this specific client (higher for regulated industries, exact pricing/claims, or strong brand guidelines)
4. Recommendation: disable, keep with monitoring, or keep with client sign-off
Account settings:
[PASTE TOGGLE STATES]Default to off, and make keeping a feature on an explicit decision, not a leftover setting
The agencies responding fastest to this are disabling Meta's AI automation across every campaign by default, then re-enabling specific features only where a client explicitly wants them and has signed off in writing. That flips the risk: instead of inheriting whatever Meta turned on, the agency controls what's active and can point to a documented decision for every exception.
Put the liability question in writing before a client asks it
Most agency-client contracts say nothing about who's responsible when a platform's own AI alters a live ad. After November 20, that silence defaults to Meta's terms, which puts the risk on the client — who will not accept that answer when they're looking at a distorted product photo that went out under their name. Get ahead of it with a short addendum instead of improvising an answer mid-crisis.
Draft a short, plain-language addendum for our client service agreements covering Meta's AI feature liability shift (effective November 20, 2026, where Meta's updated Commercial Terms make the advertiser solely responsible for anything its AI features change on an ad, including features enabled by default).
The addendum should cover:
1. That we will audit and disable Meta's AI ad features by default unless the client requests otherwise in writing
2. That any feature kept active was a documented, informed decision, not an oversight
3. What happens if Meta re-enables or ships a new auto-apply feature mid-campaign — our monitoring commitment and response time
4. That ultimate platform-level liability per Meta's terms sits with the account owner (the client), while we commit to the audit and monitoring process above
Keep it two paragraphs, plain English, no legal jargon padding.Fold the audit into every pre-launch QA, not just a one-time sweep
Exverus's ads passed quality checks before anyone noticed Meta had altered them — a one-time audit before November 20 closes the current gap but not the next one. Add a toggle-state check to the standard pre-launch checklist for every new campaign and every account handoff, so a feature that gets silently re-enabled next quarter gets caught before launch instead of after a client screenshot.
What changes
No scramble on November 20, no distorted product photo or mistranslated claim going live on a client's account without anyone having chosen it, and a contract that says in writing who's responsible when Meta's AI does something nobody asked for. The agency controls what Meta is allowed to touch instead of inheriting whatever shipped as default-on.
Meta's revised Commercial Terms add a new "AI Features" section, and the headline change is specific: advertisers are now "solely responsible" for whatever Meta's AI creates or changes on their ads. The earlier version of the terms only assigned that liability when an advertiser chose to use an AI tool, at its own discretion. The new section drops both the choice and the discretion — it applies whether or not the advertiser ever opted in. The terms were set to take effect October 30, 2026, then delayed to November 20 after publisher and advertiser complaints to EU and UK regulators about a related clause. The delay didn't change the substance of the AI liability section.
This has already happened to real advertisers
REI had an AI-altered product photo of a bicycle go live with a second set of handlebars, illegible text, and a missing crank arm, after Meta auto-enrolled the account into an AI personalization tool. Bookshop.org had an AI-illustrated image added to a live ad through an automated account update, with no request from the retailer. Agency Exverus found Meta had added sitelinks pointing to the wrong pages, inaccurately translated ad copy, and AI-generated carousels built from a client's own photos — on campaigns where the feature had never been manually turned on. In every case, the altered ad passed Meta's own quality checks before anyone on the advertiser's side noticed.
Why the "I didn't choose this" defense stopped working
Under the old terms, an advertiser could reasonably argue they never opted into whatever went wrong, because liability only attached when they chose to use the tool. The new terms remove that argument entirely. After November 20, it does not matter whether a client's account owner ever clicked anything — if Meta's AI touched the ad, the advertiser carries the risk, in Meta's own words, for results that "may not always [be] accurate, complete, or suitable."
What closes the gap
The fix isn't complicated, it's just not something most agencies have ever had a reason to do: pull the current AI feature toggle state for every client account, default everything to off, and only re-enable a feature when a client has explicitly asked for it and the decision is on record. Agencies like AdCellerant and Go Fish Digital are already recommending exactly this — systematic disablement first, selective re-enablement second. Pair that with a one-line addendum in client contracts stating who's responsible for Meta's automated changes, and the ambiguity that currently exists in almost every agency-client agreement disappears before a client ever has to ask about it.
Bottom line
November 20 isn't a date agencies can wait out — it's a date where the default behavior of an account most agencies have never audited becomes the client's legal and financial problem, and by extension, the agency's relationship problem. The audit takes half a day per batch of accounts. The alternative is finding out about a distorted photo or a mistranslated claim from a client's screenshot, the same way REI, Bookshop.org, and Exverus's clients did.