A third-party advocacy group released a fully AI-generated political video this week, sparking fresh criticism that current disclosure rules are not enough to maintain public trust as the technology becomes cheaper and more convincing. Build Canada, a non-partisan civic group backed by tech entrepreneurs, posted the one-minute video less than three weeks after the Conservative Party tested its own partially AI-generated social media clip.
The Build Canada video opens on a synthetic ribbon-cutting ceremony for an unspecified transit project. Three artificial politicians smile, shake hands, and wave to an artificial press pool. The scene then cuts to a young boy in the crowd who ages into an old man while the construction site remains untouched, a "Coming Soon" sign finally collapsing in front of it. "All levels of government in this country love to announce things, they're really great at that, but they have a really hard time following and delivering results," Build Canada CEO Lucy Hargreaves told The Hill Times. "It's a basic message."
Hargreaves, a former federal Liberal cabinet staffer, said the organization viewed the project as an experiment in a new medium and expected mixed reactions. The video drew condemnations from the left-some called the quality "slop"-and praise from the right, with Conservative Leader Pierre Poilievre and former party leader Erin O'Toole sharing it. Hargreaves declined to name the software used or the exact budget, but she estimated a comparable AI-generated video would cost between $6,000 and $10,000. Producing the same concept with real actors and sets, she said, would likely cost $250,000 to $300,000.
The cost of replacing human involvement
Digital strategists argue that the political sector is moving beyond simple AI enhancements into ethically uncharted territory. Nathaniel Arfin, founder of digital campaign firm Chester Hill Solutions and a former enterprise AI deployment expert, said generating an entire political message with synthetic humans crosses a line. "Instead of using it to skip some work or enhance some steps, you're taking away the human involvement, and it creates more distrust in what you're seeing and feeling about it," he said. "That to me is where it crosses a really, really big line."
Arfin pointed to the "Will Smith eating spaghetti" benchmark-a side-by-side comparison of AI-generated video clips from 2023 and 2025 that show how quickly the technology has improved. The progress in Generative Video Courses and open-source models means synthetic footage is now harder to spot, and Arfin said communicators need a higher ethical standard than what legislation requires. "If we don't consider the ethical framework around using these things to generate people, identifiable or not, especially in political communications, we're going to have a huge problem," he said.
Bias in the machine
Megan Buttle, founder of AI advisory firm Real Signal, said no AI system is neutral. "There's absolutely no such thing as an unbiased AI," she said. "They're all trained by individuals, and those individuals' biases shape their training, whether known or not." Even if developers could remove every trace of bias at the training stage, she explained, models continue learning from user interactions-a phenomenon she called the "mirror effect."
Buttle also questioned whether current rules around parody and satire provide adequate transparency for emotionally charged political content. Federal legislation introduced in Bill C-25, the Strong and Free Elections Act, prohibits deepfakes intended to mislead voters but carves out space for satire. Government House Leader Steven MacKinnon told a Senate committee in May, "When citizens cannot trust in what they are seeing, then we have a fundamental problem in our democracy. I don't mind being made fun of. That goes with the job. But having someone say I said something that I obviously did not say is another matter." Buttle said a "tiny AI label is not good enough to moderate" content designed to hit audiences emotionally, even when it technically qualifies as parody.
Why this matters for PR and communications professionals
Communications teams now operate in a gap between what the law allows and what audiences will tolerate. The Build Canada case shows that AI-driven production slashes video budgets by more than 95%, making it an attractive option for advocacy campaigns. But the backlash also reveals a reputational cost that disclosure alone may not offset. Hiring human talent, documenting production methods, and choosing where to stop short of full synthetic generation are fast becoming strategic differentiators. Professionals who understand the training data, bias risks, and evolving detection techniques will be better positioned to advise organizations on when AI helps and when it hurts. For teams building those skills, AI for PR & Communications Training covers both the production capabilities and the ethical frameworks needed to make those calls.
Your membership also unlocks:
