The Sources and Response to Scams Using AI and Fake News Websites

Executive Summary

A network of Facebook pages, backed by websites hosted around the world, continued advertising disinformative and deepfaked political content to Canadians during the 2025 federal election. The network grew from more than 40 identified pages to 83 in the space of a week, and these pages share characteristics with the vast global manipulation networks (an estimated 3.8 million Facebook pages worldwide) documented by a Reset Tech report, though naming conventions do not conclusively link them. Nearly 40% of advertisers disclosed no location and none self-identified as based in Canada; among those reporting a location, the United States, Ukraine, and Vietnam were most common. Linked websites were served largely from California (11 of 14 domains, via Cloudflare, from San Francisco and Los Angeles), with the rest from Frankfurt and Roosendaal, and several domains served fake news content only to visitors with Canadian internet addresses while showing innocuous content to visitors elsewhere, including when accessed via VPN from the United States, Mexico, Argentina, France, Croatia, and Hong Kong.

Although Meta slightly accelerated its moderation efforts, it took down only about half of the identified fraudulent pages, with removal rates of 41% in the morning versus 55% by the end of the day on April 24, and removed content often reappeared in a similar form, including slightly altered deepfake videos of Mark Carney republished across multiple days after supposed takedown. Tests using a consumer-grade vision detection model (Llava:7b) accurately identified 94% of ads as containing political content and 90% as containing likenesses of politicians, demonstrating that simple, computationally trivial AI methods could readily flag this content, and suggesting Meta is neglecting automated detection methods well within its capabilities.