This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
Financial experts and public figures have been targeted by sophisticated AI-generated deepfake advertisements on Meta platforms (Facebook and Instagram). These ads impersonate the individuals to promote fraudulent investment groups, leading to significant financial losses for users. Despite reporting the scams, victims and experts describe a 'whack-a-mole' scenario where fraudulent ads persist, highlighting challenges in platform moderation and the evolving nature of AI-enabled financial fraud.
Beginning in March 2025, purported deepfakes impersonating Financial Times columnist Martin Wolf were distributed via ads on Facebook and Instagram, promoting an alleged fraudulent investment group. Despite repeated takedowns, the ads have reportedly persisted under new accounts. Meta data reportedly showed at least 1,700 ads reached over 970,000 users in the EU alone. The impersonations allegedly used AI-generated visuals and voice clones to mimic Wolf's likeness and mislead users.
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
Meta's automated moderation and detection systems lacked the capability or robustness to automatically identify and prevent the dissemination of these deepfake scam advertisements.
The incident involves malicious actors using AI-generated deepfakes of financial experts to run investment scams and defraud users on social media platforms.
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 risk is primarily caused by malicious human actors intentionally deploying deepfake generation tools to defraud social media users after the AI models have been trained and deployed.
Added categories peak at 3: Substantial (Epistemic Harm, Inferred)
Sophisticated AI-generated deepfake advertisements targeted financial experts to run investment scams on Meta platforms, reaching nearly a million users and causing millions in losses. While representing an evolved capability in synthetic media fraud, the incident remains a criminal and consumer protection issue with minor direct national security implications.
The AI systems used to generate the deepfakes and the facial recognition systems used to detect them operate as tools performing specific, well-defined functions (generating synthetic media or matching faces) based on direct inputs without autonomous planning.
Risk Level 3: Limited Risk. The report describes the proliferation of 'AI-generated videos' and 'deepfakes' used to impersonate public figures, which falls under the category of AI-generated content requiring transparency obligations.
The primary technology enabling the scam was the generation of hyper-realistic deepfake videos and cloned voices of financial experts to deceive investors.
The reports provide clear, detailed, and consistent accounts of deepfake investment scams targeting specific financial experts on Meta platforms, including specific metrics on ad reach (970,000 users) and financial losses (1 million+ dollars).
None. The events clearly describe financial fraud facilitated by synthetic media.
Technology
AI Deepfake Generation
Scammers use advanced AI to create hyper-realistic video and voice clones.
Inadequate Detection Tools
Meta's AI tools failed to automatically flag and block deepfake scam ads.
Data Inputs
Internet Scraping of Public Data
Scammers scraped legitimate public images and voices to train AI models.
Fake Account Registrations
Scammers used fake profiles and credentials to bypass initial ad checks.
Human Factors
User Trust in Public Figures
Victims trusted the well-known finance experts depicted in the deepfakes.
Desire for High Returns
Low interest rates drove investors to seek unrealistically high returns.
Process and Methods
Reactive Content Moderation
Meta relied on manual reporting instead of proactive automated blocking.
Slow Ad Takedown Response
Scam ads continued to reach nearly a million users after initial reports.
Management
Profiting from Fraudulent Ads
Meta indirectly earned revenue from hosting scam advertisements.
Delayed Safety Tech Deployment
Facial recognition tools were only deployed after significant escalation.
Regulatory Environment
Safe Harbor Protections
Laws limit platform liability unless active participation is proven.
Lax Enforcement and Oversight
Lack of unified national fraud-fighting agencies in some jurisdictions.
AI system involved
Identified by automated extraction; can differ from the alleged developers above.
