This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
In 2025, approximately 5,000 Swedish investors lost a total of 500 million SEK to pump-and-dump investment scams facilitated by AI-generated deepfake videos. Fraudsters impersonated well-known financial journalists and investors on Meta platforms to lure victims into fraudulent stock schemes. The incident highlights the misuse of synthetic media to damage professional reputations and cause significant financial harm to retail investors.
At least 5,000 Swedish investors were reportedly defrauded of around 500 million SEK (~$52.5 million USD) in 2025 through purportedly AI-generated deepfake investment ads circulated on Meta platforms. Fraudsters allegedly used fabricated videos and profiles of well-known finance figures such as journalist Gabriel Mellqvist and investor Günter Mårder to promote false stock tips in pump-and-dump schemes.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
Using AI systems to gain a personal advantage over others such as through cheating, fraud, scams, blackmail or targeted manipulation of beliefs or behavior. Examples include AI-facilitated plagiarism for research or education, impersonating a trusted or fake individual for illegitimate financial benefit, or creating humiliating or sexual imagery.
Also classified as
The public likeness and professional reputation of well-known financial figures were exploited and devalued through unauthorized AI deepfakes.
Fraudsters used AI-generated deepfakes of trusted financial figures to run pump-and-dump scams, defrauding over 5,000 investors of 500 million SEK.
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 harm was caused by human fraudsters who intentionally deployed AI deepfake generation tools to deceive and defraud investors.
Added categories peak at 3: Substantial (Epistemic Harm, Indirect)
A large-scale AI-enabled financial scam in Sweden defrauded 5,000 investors of 500 million SEK using highly realistic deepfakes of trusted financial experts. While not a direct threat to physical security or sovereignty, the incident represents a substantial economic security concern, showcasing how synthetic media can be scaled to cause significant financial harm and erode trust in digital systems.
The AI systems were used as tools by fraudsters to generate synthetic video and audio of financial experts to facilitate their scam.
Limited Risk: The incident involves the creation of AI-generated content such as deepfakes, which are subject to transparency obligations under the EU AI Act.
The AI was primarily used to generate realistic deepfake videos of financial experts to deceive social media users.
The reports are consistent across multiple reputable Swedish media outlets, providing clear figures on financial losses and victims. The role of AI in generating the deepfakes is explicitly stated.
None
Technology
AI Deepfake Video Generation
AI was used to create realistic fake videos of trusted financial experts.
AI Generated Fake Profiles
Fraudsters used AI to generate highly convincing fake profiles on social media.
Data Inputs
Misuse of Public Media Footage
Scammers used existing videos of experts from known media settings to train AI.
Human Factors
Overconfidence in Social Media Ads
Users trusted ads featuring well-known financial figures on Facebook.
Lack of Digital Literacy in Elders
Older generations struggled to critically assess AI-generated video materials.
Process and Methods
Inadequate Ad Moderation on Meta
Meta failed to detect and block deepfake scam ads on its platform.
Use of Private WhatsApp Groups
Fraudsters moved victims to WhatsApp to coordinate pump and dump schemes.
Management
Lack of Accountability by Meta
Meta management did not take responsibility or act against fake profiles.
AI system involved
An AI system was involved, but the reports do not identify it.
Identified by automated extraction; can differ from the alleged developers above.
