REI, the outdoor retailer with 181 stores across the United States, ran an Instagram ad last month showing a bicycle with two sets of handlebars. The image was nonsensical. REI never approved it.
The company traced the problem to Meta's AI ad tools, which had auto-enrolled REI in an AI personalization feature that altered a vendor-supplied photograph without the brand's knowledge. The original image came from a Van Rysel photo shoot featuring cyclist Amity Rockwell. Somewhere between the brand and the feed, Meta's AI decided the bike needed a second handlebar. Van Rysel confirmed the original image was unaltered. REI called the alteration "inaccurate and inappropriate" and immediately unenrolled from the tool.
Meta's response was blunt: its terms of service state that "AI can make mistakes and that it is the advertiser's responsibility to review the AI outputs." The company says its AI image generation feature is turned off by default. But the distinction between "off by default" and "accidentally switched on by a bug" is precisely where the damage sits.
The REI incident was not isolated. Business Insider reported that Karissa Tuccio, executive director of social and influencer at agency Mediassociates, said a bug that toggled AI settings on had "regularly affected most of the 15 clients" she manages on Meta's ad platform. Other advertisers described AI-generated images featuring models with legs bent the wrong way, an unrealistic grandmother in loungewear, and entirely changed product photography. One small bookshop, Quite Literally Books, discovered AI-generated distortions in its ads and requested a refund from Meta. As of publication, the refund had not arrived. Meta told the bookshop the issue was a "sporadic" and "one-off occurrence" and separately called it a "glitch."
Starting last month, Meta began automatically applying an "AI info label" to ads that use its AI tools — or third-party tools like Midjourney or DALL-E — to create or significantly edit visuals. Google added similar labels last week. The direction across platforms is clear: AI is being inserted between the brand and the audience, and the speed of adoption is outpacing the quality controls.
This matters beyond Meta. Every major ad platform is pushing AI creative tools. The question that remains unanswered is who bears the real cost when the AI gets it wrong — not legally, where the terms of service are explicit, but commercially, where the brand damage is immediate and the refund process is slow or nonexistent.
Three things worth noting:
Audit your AI settings after every campaign launch. Meta's AI features can toggle on through bugs. Tuccio's experience across 15 clients suggests this is systemic, not rare. Check the creative settings screen before any campaign goes live, and check again a day after launch.
The terms of service put you on the hook. Meta's legal position is unambiguous: you are responsible for reviewing AI outputs. If a distorted image runs and damages your brand, Meta's stance is that you should have caught it. Build a manual review step into every campaign workflow, regardless of what the default settings claim.
Save the original creative files separately. When AI tools alter images mid-campaign, you need a clean reference copy to prove what you submitted and what changed. This is especially important for product images where accuracy is not optional — and for any refund dispute that follows.
The platforms are building the tools. The platforms are shifting the liability. The advertiser sitting in the middle had better be checking every setting, every time.