The most alarming part of a viral trade-war warning was not what Donald Trump supposedly demanded. It was that the demand never existed. A new investigation by McGill University’s Centre for Media, Technology and Democracy found that a fabricated 48-hour U.S. ultimatum, paired with an invented Canadian defence-contract cancellation, amassed 5.1 million video plays across TikTok and Instagram.
The clips were part of a much larger network using real people’s faces and artificial intelligence to manufacture political commentary. After the findings emerged, TikTok banned accounts implicated in the operation, and Meta confirmed removals on Instagram. For Canadians already navigating genuine tariff disputes and economic uncertainty, the case shows how easily a convincing-looking clip can turn a false claim into a heated national argument.
The 48-Hour Ultimatum That Never Happened
The central claim sounded tailor-made for an anxious news cycle: U.S. President Donald Trump had supposedly given Canada 48 hours to back down in a tariff dispute. In the same manufactured storyline, Prime Minister Mark Carney answered by cancelling $8 billion in U.S. defence contracts. Researchers at McGill University’s Centre for Media, Technology and Democracy concluded that neither event occurred. The story was fiction packaged with the urgency of breaking diplomacy, combining two recognizable political leaders, a countdown and an enormous dollar figure into a startling social-media narrative.
Researchers traced that narrative through 161 videos distributed by 129 accounts, which collectively generated 5.1 million plays. That figure measures video plays, not five million distinct viewers, and it reflects repeated uploads of the same false storyline rather than a single clip. The distinction matters because repetition can create the impression that one source is confirming what another has already reported. Someone scrolling past different versions might assume the dispute had escalated into an official confrontation, especially when the videos sounded confident. In reality, the number of times a claim appears on a feed says nothing about whether the underlying event happened.
Almost 300 Accounts Worked From the Same Playbook
The researchers did not find a handful of unrelated prank videos. They documented at least 293 connected accounts: 257 on TikTok and 36 on Instagram. Between August 7 and November 4, 2026, those accounts published 3,201 videos that accumulated 22.1 million plays. TikTok accounted for 82 per cent of the plays and Instagram for the remaining 18 per cent. These figures represent activity the investigators could identify, not a count of unique individuals reached. The research team also cautioned that the operation was still growing when it stopped collecting data.
Its pace accelerated dramatically. The network expanded from three active accounts posting two videos a day to 157 active accounts posting 108 videos a day. Investigators linked profiles through recurring video formats, captions, titles and even matching typographical mistakes. A repeated typo can be a more revealing sign of coordination than a shared political viewpoint. To a casual viewer, the posts could look like independent reactions from dozens of people. Seen together, their similarities exposed a distribution system able to fill social feeds with apparently separate voices while recycling messages from the same source material.
Real Creators’ Faces Were Used to Deliver Invented Messages
One example began with an ordinary video that had nothing to do with geopolitics. On June 17, 2026, American creator Ashley Hetherington posted about dating and her Christian faith. On October 25, an Instagram profile called Raven published a manipulated version in which her face appeared to discuss Canadian job losses, tariffs and Alberta’s referendum. McGill’s researchers compared the original and altered footage frame by frame: her hair, eyes and head movements matched closely, while her mouth and the spoken words had changed. The account effectively borrowed the credibility of someone who had not delivered that political commentary.
The operators had extracted just 21 seconds of her original recording and looped that footage, forward and backward, through a much longer video. Another creator, financial commentator Andrei Jikh, appeared in six reels built from the same source clip but given different scripts. Investigators found that 283 of the 293 accounts relied on footage of a single real person. Across the network, they counted 226 distinct individuals and identified 104 of them. The potential harm goes beyond a misleading headline: viewers can remember a familiar face alongside words that person never spoke. It is a form of impersonation that makes an invented message feel personal, even without creating a face from scratch.
Canada’s Trade War Became the Network’s Biggest Attraction
The operation was built around subjects Canadians were already discussing. Researchers found that 84 per cent of its videos concerned Canada, while trade-war topics—tariffs, negotiations, boycotts and their economic fallout—accounted for 54 per cent of the posts and 71 per cent of their plays. That concentration helps explain why a fictional Trump deadline found such a large audience. It did not arrive in a vacuum; it was surrounded by coverage of actual tensions between Canada and the United States, making an alarming new claim easier to mistake for the latest development.
The messaging was not consistently pro-Trump or pro-Carney. Among videos about the trade war, attacks on Carney’s government were the most common single stance, appearing in 56 per cent of those posts, but individual accounts often carried arguments from both sides. Some narrators even presented themselves as Canadians in one clip and Americans in another. Other content addressed jobs, housing, interest rates and Alberta’s referendum. By shifting targets, the network could reach people with very different political loyalties. What linked the videos was less a coherent position than a willingness to turn economic and political anxieties into attention-grabbing narratives.
Repeated Videos Made False Claims Look More Credible
The researchers found a production system that did not need thousands of original ideas. Across 3,201 videos, they identified 1,123 distinct scripts. Of those, 326 appeared more than once, while another 142 were rewrites of earlier scripts that made essentially the same argument in different words. One story could therefore turn up under different account names, with different-looking speakers and slightly adjusted wording. The repeated fake ultimatum was the most striking example, but the wider pattern showed how little new material was necessary to create a crowded-looking conversation.
The team also noticed a relationship between early popularity and later reuse. The first upload of a script that would eventually be reposted attracted a median of 1,626 plays; scripts uploaded only once had a median of 664. Those figures show an association, not proof that popularity alone dictated the operators’ choices. Still, the pattern is consistent with recycling material that attracts attention. In everyday terms, a story that sparks arguments in one corner of TikTok can quickly be repackaged as several apparently independent reactions. For audiences, encountering the same claim repeatedly can feel like corroboration when it may simply reflect copying.
Only a Small Share of Commenters Recognized the Deception
The comments revealed how convincing the artificial speakers could be. In the TikTok discussions researchers sampled, only 6.3 per cent of comments suggested that a video was fake, AI-generated, copied or propaganda. Most commenters either agreed with the message or challenged its argument without questioning who was supposedly speaking. Roughly one in ten addressed the person on screen directly, sometimes praising them or asking them to check their facts. The researchers examined comments from videos associated with 97 confirmed TikTok accounts, rather than claiming to measure the reactions of every viewer.
Audience responses also rewarded the illusion. Comments addressing the narrator as a real person received 14,713 likes in the researchers’ tally, compared with 4,241 for comments doubting a video’s authenticity. None of the 100 most-liked comments in their analysis questioned whether the content was genuine. This does not establish how many viewers believed the claim or changed their political opinions. It does reveal something consequential about the conversations taking place beneath the clips: people were debating the fabricated statements while the impersonation itself largely escaped notice. The apparent speaker had become part of the story, even though that speaker was not truly speaking.
AI Detection Systems Failed to Flag Most Manipulated Videos
The network exposed a weakness in automated AI warnings. In a sample of 733 TikTok videos, researchers found that the platform had labelled 22 of 24 clips featuring completely synthetic avatars. But only 26 of 709 clips made from real people’s footage carried TikTok’s AI label—about four per cent. None of those 733 videos had been disclosed as AI-generated by the accounts that posted them. The contrast suggests that a familiar human face, even with digitally altered lips and narration, can be harder for a labelling system to identify than an entirely artificial character.
The investigators also used detection software that rated the narration as synthetic in 94 per cent of the network’s videos. A separate text detector estimated that 46 per cent of scripts were entirely machine-written; such scores are estimates, not definitive proof of authorship for any particular script. TikTok’s published rules require labels for realistic AI content, including videos that make real people appear to say things they never said. The danger is assuming the absence of a label means a video is genuine. Here, technology created much of the deception, while a different layer of technology frequently failed to alert viewers to it.
Investigators Found Overseas Connections but No Confirmed Operator
Researchers found strong signs that the operation was being managed outside North America, but they could not identify its operators. Just one per cent of videos were posted during the six-hour period from 1 p.m. to 7 p.m. Eastern Standard Time, which corresponds to overnight hours in parts of East and Southeast Asia. Of 255 TikTok accounts created in 2026, 80 per cent were registered between midnight and 8 a.m. Eastern. Twelve of 184 accounts with readable location or language data had Chinese-language settings or a registration in China. Those are clues about possible operational ties, not proof of a Chinese government role.
The network also appeared disconnected from events it claimed to understand. For two weeks after Alberta’s October 19 referendum, researchers still found 43 videos across 37 accounts describing the vote as something yet to happen. Some profiles simultaneously discussed its result. Such mismatches fit a system distributing recycled scripts without consistently checking local developments. Investigators found no clear advertising, sales links or sponsorships explaining how the accounts made money. They raised several possibilities, including audience-building for later use or a paid influence operation, but did not establish a motive. The uncertainty matters: a suspicious posting schedule is evidence worth investigating, not a licence to name a perpetrator without proof.
TikTok and Instagram Removed Accounts Following the Investigation
The response became public after the McGill report appeared on November 12. The researchers said every account they had identified was still active as of November 11. On November 13, TikTok told The Canadian Press it had permanently banned the accounts named in the investigation. The company said its own follow-up inquiry also uncovered additional accounts and content that broke its rules on deceptive behaviour and misleading AI-generated material. That action directly targeted the distribution network, rather than merely adding warnings to a handful of popular videos.
Instagram’s response became clearer later the same day. Although Meta initially declined to comment on the findings, a spokesperson subsequently confirmed to The Canadian Press that the accounts identified in the report had been removed from Instagram. The statements establish that both platforms acted, but they do not provide an independent accounting of every repost or the precise timing of each removal. For the creators whose footage was appropriated, taking down an account can limit further misuse; it cannot undo the impressions already left on viewers. For platforms, the episode raises a difficult question about why a large network remained visible until outside researchers connected the evidence.
Canada Faces a Bigger Challenge in Protecting Public Trust
The case reaches beyond one fictional Trump deadline. Researchers counted 42,538 followers across the 257 TikTok accounts, including 16,055 accounts registered in Canada. Those followers were not all necessarily fooled, and the investigation did not measure changes in voting behaviour or public attitudes. Still, the figures show how artificial political speakers can enter ordinary recommendation feeds and acquire an audience before their origins are understood. Canada’s October 2026 national security strategy promised to strengthen the country’s ability to detect foreign information manipulation using open-source information, placing the broader issue squarely within national security planning.
For the public, the Canadian Centre for Cyber Security recommends treating emotionally charged claims, sensational headlines and manipulated imagery as reasons to verify a story rather than immediately share it. An alleged 48-hour diplomatic ultimatum should be checked against official government communications and credible reporting, not accepted because a familiar face repeats it on video. For platforms, researchers and policymakers, the challenge is more structural: detecting coordination, protecting people from impersonation and making reliable evidence available before a fiction acquires millions of plays. Removing accounts is important, but the lasting test is whether Canadians can distinguish genuine developments from manufactured confrontations while the next major news event is still unfolding.
