Des vidéos hypertrucées du premier ministre Carney utilisées pour frauder des Canadiens et perturber l'élection fédérale de 2025

During and after the April 2025 Canadian federal election, an extensive AI-enabled disinformation and fraud campaign targeted Canadians using deepfake videos of Prime Minister Mark Carney, CBC journalist Rosemary Barton, CTV news anchors, and Elon Musk.

The campaign had two dimensions. First, a viral deepfake video — created using Fish Audio, a free AI voice cloning tool — falsely depicted Carney announcing that the government would ban vehicles manufactured before 2000. The video, which manipulated authentic footage from a March 27 press conference on tariff response, appeared on TikTok around the time of the April 28 election — the DFRLab reported it was first published directly before the vote, though independent fact-checkers found the earliest archived posts dated to May 3–4 — and reached millions of views (DFRLab, 2025), spreading to X where a single repost garnered 2.4 million views, with at least 18 posts sharing the video across the platform (DFRLab, 2025). Although TikTok labeled the video as AI-generated, it continued to be amplified by influencers even after removal (DFRLab, 2025).

Second, a sophisticated network of over 40 Facebook pages and 25+ accounts — managed by operators traced to Ukraine, Indonesia, the United States, Angola, Romania, and Vietnam — ran AI-generated deepfake "news segments" featuring Carney, Barton, and Canadian news anchors to funnel victims into fraudulent cryptocurrency investment platforms (CBC News, 2025; Canadian Digital Media Research Network, 2025). The scam operated through a multi-step funnel: Facebook ads pushed deepfake CBC and CTV reports to fake news websites, where Canadians were invited to provide contact information and invest a minimum of approximately $350 on platforms with rotating names — CanFirst, QuilCapital, Quantum AI, TokenTact, and others (CBC News, 2025; Canadian Digital Media Research Network, 2025). "Financial advisers" would then pressure victims for larger investments, sometimes depositing small "profits" to build trust before extracting larger sums (CBC News, 2025).

A 70-year-old retiree from Prince Albert, Saskatchewan lost approximately $2,800 after encountering what appeared to be a CBC News interview with Rosemary Barton and Mark Carney promoting a government-backed crypto opportunity (CBC News, 2025). Saskatchewan's Financial and Consumer Affairs Authority issued at least four separate investor alerts between June and September 2025 (Saskatchewan Financial and Consumer Affairs Authority, 2025). The Regina Police Service reported that losses from the QuilCapital scam alone were expected to exceed $1 million (Saskatchewan Financial and Consumer Affairs Authority, 2025).

The Canadian Digital Media Research Network identified Meta's Canadian news ban under the Online News Act as a significant contributing factor (Canadian Digital Media Research Network, 2025). The CDMRN argued that the ban removed legitimate news content from Facebook and Instagram, creating conditions where AI-generated fake news content faced no competition from real journalism and appeared authoritative to users unfamiliar with the ban's effects (Canadian Digital Media Research Network, 2025). Meta removed pages and accounts when flagged by CBC and researchers, but only approximately half of identified scam pages were taken down, and new ones were created daily (CBC News, 2025). The CDMRN also noted that Meta's January 2025 decision to end its fact-checking programs further reduced the platform's capacity to address false content (Canadian Digital Media Research Network, 2025).

A preprint study by researchers at the Université de Montréal and Mila, analyzing 187,778 social media posts during the election period, found that 5.86% of election-related images were flagged as deepfakes, with right-leaning users posting flagged images at a higher rate than left-leaning users (arXiv, 2025). While most deepfakes were benign memes rather than deliberate misinformation, the study confirmed that realistic fabricated images drew higher engagement (arXiv, 2025). No criminal charges related to the scam operation have been reported.