This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
A Southampton man was targeted over five months by scammers using AI-generated deepfake videos, images, and voice clips of celebrities to conduct romance and prize-based phishing scams. The victim was manipulated into sending £200 in non-refundable gift cards after being convinced he was in a romantic relationship with a celebrity. The incident highlights a broader trend of using generative AI to create convincing synthetic identities for financial fraud.
Over about five months in 2025, Paul Davis, a Southampton, UK man, reports that he was repeatedly targeted by scammers using purported deepfake videos and images of celebrities including Jennifer Aniston, Mark Zuckerberg, Elon Musk, and Ellie Goulding. The perpetrators allegedly rotated personas to sustain a romance and prize scam, extracting £200 in gift cards. This case suggests a new shift from one-off celebrity deepfakes to persistent, multi-persona targeting of a single vulnerable victim.
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.
The incident involves scammers using AI-generated deepfakes to impersonate celebrities and conduct a targeted romance and financial fraud scam for personal gain.
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 the intentional actions of human scammers who deployed readily available AI deepfake tools to defraud the victim.
Added categories peak at 3: Substantial (Psychological Harm, Inferred)
This incident involves the use of generative AI tools to conduct romantic and financial scams against individuals. While causing personal distress and financial loss, the national security impact remains minor as it lacks state-sponsored coordination, systemic economic threats, or target-rich intelligence manipulation.
The AI systems involved functioned strictly as tools. Scammers used readily available generative AI software to produce synthetic videos, images, and voice clips based on direct inputs, without the AI exhibiting autonomous decision-making or goal-directed planning.
Limited Risk: The incident involves the use of AI-generated content such as deepfakes, which are classified as limited risk under the EU AI Act and are subject to specific transparency and disclosure obligations.
The primary function of the AI systems used in this incident was to generate realistic deepfake videos and synthetic voice clips of celebrities to deceive the victim.
The reports provide consistent, detailed accounts of the scam targeting Paul Davis, including the specific amount lost, the platforms used, and the types of AI-generated media involved. Multiple sources corroborate the core facts of the incident.
None.
Technology
Highly Convincing Deepfakes
AI-generated videos and cloned voices mimic celebrities with high realism.
Accessible Generative AI Tools
Free and easy-to-use AI tools allow scammers to easily fabricate media.
Neural Network Image Generation
AI models generate customized fake certificates and prize notifications.
Data Inputs
Publicly Available Celebrity Media
Scammers use public photos and voice samples to train AI models.
Victim Personal Data Harvesting
Scammers scrape names and details from social media to refine AI outputs.
Lack of Authenticated ID Inputs
Absence of secure, verified digital identity credentials on platforms.
Human Factors
Vulnerability Due to Depression
Victim isolation and mental health struggles made him susceptible.
Parasocial Celebrity Bonds
Victims form one-sided emotional trust with famous public figures.
Limited Technological Literacy
Difficulty distinguishing AI-generated content from real interactions.
Process and Methods
Slow Platform Moderation Actions
Social media platforms are slow or ineffective at blocking fake accounts.
Unregulated Gift Card Payments
Use of non-refundable, untraceable payment methods like Apple gift cards.
Inadequate Verification Processes
Lack of robust real-time verification for celebrity profiles.
Management
Inadequate Platform Safeguards
Facebook failed to proactively prevent automated deepfake distribution.
Flawed Reporting Mechanisms
Platform handles user reports piecemeal without systematic resolution.
Neglecting Vulnerable User Risks
Failure to design safety features protecting vulnerable users from scams.
Regulatory Environment
Lagging Cybercrime Legislation
Laws struggle to keep up with rapid generative AI advancements.
Jurisdictional Enforcement Barriers
Scammers operate internationally, hindering law enforcement prosecution.
Lack of Platform Liability Rules
Absence of strict regulations holding platforms liable for hosting scams.
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.
