Meta's most sensitive safety boundary was hit by two forces at once: AI-generated abuse imagery and paid advertising.
WIRED reported that Tech Transparency Project researchers found more than 50 offending image and video ads in Meta's ad library. They appeared across Facebook, Instagram, Messenger, and Threads, and some linked to nudify or face-swapping apps.
The first gate failed
Meta's ad library showed ads reaching the US, UK, and more than a dozen European countries. At least one reached 2,563 accounts in Europe. Some had remained visible in the ad library for months.
“These ads made no effort to mask the images or hide what they were promoting”
Meta said the ads were removed and pointed to broader enforcement work, including 36 million pieces of child exploitation content removed last year.
The chain runs beyond Meta
WIRED said some disclosed advertisers were tied to Chinese companies. Several ads linked to MaskAI; Apple removed that app from the App Store after WIRED asked about it. The News Minute also recorded Meta's July response to an India case, including 160,000 suspicious accounts removed in India over six months.
The real test is repeat prevention
The issue is not only model output. Bad actors can automate image generation, account creation, landing pages, and re-posted ads, while platforms use automated review to catch them. For ad systems, the bar should be higher than ordinary user posts because payment and targeting create extra responsibility.
Sources: WIRED, Meta official statement, The News Minute, arXiv:2607.05407, CocoLoop; verification covers 50+ offending ads, 2,563 European reached accounts, 30 additional samples, 36 million removals, 160,000 India suspicious accounts, and 15 AI-CSAM safety problems.