Meta Ran Ads for AI App That Nudifies Female Politicians

Meta's advertising network recently hosted promotional campaigns for Kromix, an AI application that generates sexually explicit deepfakes, including videos depicting female U.S. politicians. Discovered by the Tech Transparency Project, the campaign consisted of 32 ads targeted exclusively at male users under the tagline "Unleash Next-gen AI Magic." Although most of these ads received 10 or fewer impressions during their five to 46 hours of activity, their approval highlights ongoing vulnerabilities in Meta's automated ad-screening systems. The Kromix app, which has since been removed from Apple's App Store, allowed paid users to upload photos to generate explicit content.

To bypass Meta's automated filters, the advertisers employed adversarial tactics, such as using benign cover photos of landscapes like beaches and mountains to mask the underlying explicit video content. Meta's head of women's safety, Cindy Southworth, acknowledged these evasion techniques, noting that the company aggressively targets sexual exploitation. Between November of last year and January, Meta removed 344,000 ads promoting nudify apps. However, critics point out that Meta's heavy reliance on automated AI tools for ad vetting, combined with recent layoffs of human moderators, leaves the platform highly vulnerable to bad actors who continuously adapt their methods.

This incident is part of a broader pattern of moderation failures. The Tech Transparency Project also identified over 50 ads on Meta's platforms promoting MaskAI, another face-swapping app that was subsequently pulled from the App Store. For AI developers and trust and safety practitioners, this development underscores the limitations of relying solely on automated classifiers to police generative AI outputs and advertisements. As malicious actors increasingly use benign wrappers and adversarial framing to bypass safety filters, platform developers must design multi-layered moderation pipelines. Relying on automated systems without robust human-in-the-loop oversight remains a significant operational risk for major ad networks.