This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
A network of 20 inauthentic YouTube channels has been identified using generative AI to create and distribute political disinformation targeting Albertan audiences. The content, described as 'slopaganda', promotes Alberta secession and U.S. annexation, accumulating nearly 40 million views. Researchers warn that this coordinated effort exploits political divisions and undermines Canada's democratic integrity, calling for greater platform transparency and regulatory oversight.
Researchers reported that a network of 20 inauthentic YouTube channels used purportedly AI-generated avatars and voiceovers with deepfake-style thumbnails, as well as paid voice actors, to pose as Albertan commentators and promote false or misleading claims about Alberta secession and U.S. annexation. The channels allegedly drew nearly 40 million views and distorted political debate, reportedly undermining Canada's information ecosystem.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
Using AI systems to conduct large-scale disinformation campaigns, malicious surveillance, or targeted and sophisticated automated censorship and propaganda, with the aim of manipulating political processes, public opinion, and behavior.
Also classified as
The videos contain frequent and obvious lies, fabricated political deals, and fake polls designed to mislead viewers.
The massive volume of AI-generated 'slopaganda' threatens to pollute the online information ecosystem and manipulate search engine results.
The incident involves a coordinated, large-scale campaign using AI-generated content to spread political disinformation and manipulate public opinion regarding the Alberta secession referendum.
Human
Due to a decision or action made by humans
Intentional
Due to an expected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed
The incident was driven by human actors who intentionally deployed generative AI tools post-deployment to run a coordinated political influence campaign.
Added categories peak at 4: Severe (Epistemic Harm, Direct)
A coordinated campaign used generative AI to produce and distribute political disinformation ('slopaganda') targeting Albertan audiences. Accumulating 40 million views, the campaign promoted secession and annexation, threatening Canadian democratic integrity, societal stability, and national sovereignty ahead of an independence referendum.
The generative AI systems were used as tools by human operators to perform specific content generation functions, such as creating voiceovers, scripts, and video avatars, without exhibiting independent goal-directed planning or autonomous decision-making.
Limited Risk: The report describes the use of AI-generated content (deepfakes, synthetic voiceovers, and avatars), which falls under the transparency obligations of Limited Risk systems to ensure users are aware they are interacting with AI-generated media.
The AI systems were used to generate scripts, voiceovers, and video avatars specifically for distribution on social media platforms like YouTube.
The reports provide consistent and detailed findings from multiple research groups (CDMRN, MEO, Insikt Group) regarding the scale, tactics, and AI technologies used in the influence campaign. The role of generative AI in creating the 'slopaganda' is explicitly documented, and the platform metrics (40 million views) are clearly stated.
The campaign could be interpreted purely as a profit-driven 'spam' operation to generate ad revenue rather than a politically motivated foreign influence campaign.
Technology
Generative AI Tool Accessibility
Generative AI allows cheap, mass production of political slopaganda.
AI Voice and Avatar Generation
AI voiceovers and avatars mimic authentic local presenters to mislead.
Algorithmic Amplification
Recommendation algorithms prioritize highly engaging, divisive AI content.
Data Inputs
Out-of-Context Media Inputs
Old news clips are stripped of context to train and generate AI scripts.
AI Training Data Poisoning
AI slop is uploaded to manipulate future LLM search results on secession.
Fabricated Information Inputs
Fake polls and lies are used as data inputs for generating AI scripts.
Human Factors
Cognitive Vulnerability
Audiences easily share AI slop that aligns with pre-existing grievances.
Inauthentic Local Performance
AI allows foreign actors to pretend to be passionate local Albertans.
Lack of Audience Verification
Viewers share AI content without verifying facts or noticing errors.
Process and Methods
Inadequate AI Moderation
YouTube processes fail to quickly detect and remove coordinated AI networks.
Lack of Metadata Transparency
No public tracking of account creation, ad targeting, or geographic data.
Absence of Community Notes
Lack of crowd-sourced flagging allows AI slop to spread unchecked.
Management
Prioritizing Engagement Profit
Platforms focus on views and ad revenue over policing AI-driven campaigns.
Inadequate Risk Mitigation
Platform management failed to anticipate AI-enabled covert operations.
Restricted Researcher Access
Management denies researchers API access to identify coordinated AI networks.
Regulatory Environment
Lagging AI Regulations
Governments lack frameworks to regulate AI-generated political disinformation.
Limited Electoral Powers
Elections Alberta historically lacked power to address online deepfakes.
Reactive National Security
Institutional responses move slower than algorithmic AI threats.
AI system involved
Identified by automated extraction; can differ from the alleged developers above.
