Tbilisi-Based Call Center Allegedly Uses AI-Driven Scripts to Defraud Over 6,000 Victims of $35 Million

A large-scale AI-assisted financial scam, allegedly operated from Tbilisi, Georgia, used deepfake celebrity endorsements and manipulated victims through fraudulent trading dashboards that simulated high returns. Call center agents, trained with AI-driven persuasion tactics, convinced individuals to invest more money while falsely promising profits. Many reportedly lost their life savings, with the total losses reaching $35M across more than 6,000 victims.

Risk classification

• Primary risk domain: 4 Malicious actors
• Primary risk subdomain: 4.3 Fraud, scams, and targeted manipulation

The incident involves malicious actors using AI-generated deepfakes, cloned websites, and fake documents to conduct targeted financial fraud and scams against individual victims for personal gain.

Additional risk subdomains

• 3.1 False or misleading information: The scammers deployed fake trading platforms displaying simulated live market data and fabricated account balances to mislead victims.
• 2.1 Compromise of privacy by obtaining, leaking or correctly inferring sensitive information: Scammers obtained sensitive personal information and remote access to victims' devices using AnyDesk under false pretenses.

Causal factors

• Entity: Human
• Intent: Intentional
• Timing: Post-deployment

The financial scam was orchestrated intentionally by human malicious actors who deployed AI tools post-deployment to generate deepfakes and fraudulent documents to deceive victims.

EU AI Act risk tier

• Risk tier: 3 Limited Risk

Risk Level 3: Limited Risk. The incident involves 'AI-generated content such as deepfakes' used to impersonate celebrities and 'cloned website interfaces' to deceive users, which falls under transparency obligations for synthetic content.

AI system and alleged parties

• AI system: unspecified
• AI purpose: Deepfake Video Generation; Deepfake Image Generation
• Behaviour type: Tool
• Alleged developer: Unknown deepfake technology developers, Third-Party AI Service Providers for Call Center Training, Fraudulent Trading Platform Developers
• Alleged deployer: Scammer Operatives in Georgia, Bulgaria, Cyprus, and Spain, Meri Shotadze, Golden Currencies, Akaki Kevkhishvili, AK Group, AdmiralsFX
• Alleged harmed parties: Small business owners, Retirees seeking investment opportunities, People with neurological disorders, Individuals deceived by celebrity deepfake impersonations, Financial professionals, Financial fraud victims, Elderly investors

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Substantial. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

• Physical: direct Negligible, indirect Negligible
• Infrastructure: direct Negligible, indirect Negligible
• Property: direct Negligible, indirect Negligible
• Financial: direct Substantial, indirect Negligible
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Minor, indirect Negligible
• Differential treatment: direct Minor, indirect Negligible
• Civil rights: direct Negligible, indirect Negligible
• Democracy: direct Negligible, indirect Negligible
• Privacy: direct Substantial, indirect Substantial
• Psychological: direct Negligible, indirect Substantial
• Epistemic: direct Minor, indirect Substantial
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Financial

Reported: The report explicitly describes massive financial losses, with the Georgian operation stealing $35.3 million from over 6,000 victims, averaging approximately $5,713 per victim.

Directly caused: Average financial loss per occurrence is difficult to separate as it was a single continuous operation, but individual victims lost significant sums, such as Mark losing £27,000 and another victim losing £162,000.

Inferred additional harm: The total financial loss across all operations mentioned (including the Israeli/European network and Spanish scam) is inferred to exceed $300 million.

Malicious content

Reported: The report explicitly describes the creation and spread of malicious content, specifically deepfake videos and cloned websites.

Directly caused: Scammers created deepfake videos of celebrities like Ben Fogle and Martin Lewis to promote fraudulent crypto schemes on social media platforms like Facebook.

Inferred additional harm: It is likely that thousands of other deepfake ads and cloned websites were generated and distributed across social media platforms to target potential victims.

Differential treatment

Reported: The report explicitly describes differential targeting of individuals based on demographic profiles.

Directly caused: Scammers used algorithms to select victims whose profiles matched their criteria, specifically targeting wealthy nations like the UK and Canada to maximize financial gain.

Privacy

Reported: The report explicitly describes privacy violations where scammers obtained personal data and remote access to victims' devices.

Directly caused: Scammers used remote access software like AnyDesk to gain control of victims' phones and computers to bypass bank security.

Indirectly caused: Victims' personal contact details and financial information were harvested and stored in internal databases by the scam network.

Inferred additional harm: It is likely that the harvested personal data of the 6,000+ victims was shared or sold within cybercriminal networks, exposing them to further privacy risks.

Psychological

Reported: The report explicitly describes psychological distress and mental-health impacts on victims of the scam.

Indirectly caused: Victims experienced severe distress, anxiety, and 'dark thoughts' after losing their life savings, such as a 61-year-old woman left wrestling with suicidal ideation.

Inferred additional harm: It is highly likely that many of the other 6,000+ victims experienced significant psychological trauma, anxiety, and depression due to the sudden loss of their life savings.

Epistemic

Reported: The report explicitly describes epistemic harm through the creation of highly realistic fake trading platforms and deepfake news articles.

Directly caused: Scammers deployed fake trading software showing simulated live market tickers and fabricated account balances to convince victims they were making real profits.

Indirectly caused: The proliferation of deepfake celebrity endorsements on social media eroded public trust in online information and media authenticity.

Inferred additional harm: The widespread use of deepfakes likely contributed to a broader societal erosion of trust in digital media and financial advertisements.

People affected

• Occurrences reported: 1
• People reportedly harmed: 6179
• People reportedly exposed: 6179

Potential causes

Management

• Prioritizing Ad Revenue: Social media platforms hosted fake ads without adequate safety verification.
• Lack of Proactive AI Risk Audit: Financial institutions failed to anticipate AI-enabled identity fraud.

Technology

• Generative AI Deepfakes: Scammers generated realistic celebrity videos and voices to lure victims.
• AI Passport Image Generation: ChatGPT used to easily generate fake passport images to bypass bank KYC.
• Rapid AI Web Cloning Tools: AI tools allowed scammers to quickly copy real trading platforms like Exante.

Data Inputs

• Targeting Profile Data: User profile data was harvested by algorithms to target vulnerable victims.

Human Factors

• Overreliance on Digital Content: Victims trusted highly realistic deepfakes of trusted public figures.
• Coerced Compliance Deception: Victims lied to bank compliance staff under direct instruction from scammers.
• Remote Device Control Consent: Victims handed over remote PC control via AnyDesk due to tech struggles.

Process and Methods

• Weak KYC Verification: Challenger banks accepted AI-generated fake passports to open accounts.
• Ineffective Ad Moderation: Social media platforms failed to filter out AI deepfake promotional ads.
• Structured Boiler Room Tactics: Organized multi-tier call centers systematically groomed and pressured victims.

Regulatory Environment

• Lagging AI Fraud Regulations: Regulators have not fully addressed AI-driven fraud and cyberthreats.
• Jurisdictional Enforcement Gaps: Cross-border operations in Tbilisi evade UK and US law enforcement.
• Inadequate Big Tech Oversight: Lack of government mandates to stop platforms publishing dangerous fakes.

Information quality

• Classification confidence: High
• Reason for confidence: The reports are highly detailed, drawing from a massive data leak of over 1 million recorded calls, internal spreadsheets, and regulatory warnings. The roles of deepfakes, cloned websites, and fake documents are explicitly documented.
• Ambiguities identified: The specific AI models used to generate the deepfake videos are not named, though ChatGPT is explicitly mentioned for passport image generation.
• Alternative interpretations: None. The incident is clearly a coordinated financial fraud facilitated by generative AI tools.

A highly sophisticated financial fraud operation (AK Group) utilized generative AI, including deepfakes and synthetic document generation, to defraud over 6,000 victims of $35 million. While the incident represents a significant evolution in cyber-enabled financial crime and exposes vulnerabilities in digital banking security, its direct national security impact remains minor.

• Overall national security impact: Minor
• Scope: Multiple nations
• Primary target: UK
• Other affected: Canada, Georgia, Spain, Bulgaria, Cyprus, Israel
• Alleged perpetrator: AK Group

Threat characteristics

• Imminence: Long-term. Represents an ongoing strategic cyber-fraud capability and regulatory challenge rather than an active, imminent national security crisis requiring immediate intervention.
• Autonomy: Human-controlled. AI systems were utilized strictly as tools by human operators who directed the creation of fraudulent assets and managed the execution of the scam.
• Novelty: Evolved capability. Represents a significant advancement in fraud operations by integrating generative AI tools to rapidly scale deepfakes, cloned assets, and synthetic identity documents.

Impact by dimension

• Physical security: Negligible. The incident was entirely virtual and financial, with no impact on physical systems, critical infrastructure, or human physical safety.
• Information security: Minor. Utilized deepfakes of public figures and cloned websites, but the intent was financial fraud rather than systematic state-sponsored information warfare or intelligence compromise.
• Sovereignty: Negligible. No significant compromise of core government operations, electoral systems, or state sovereignty occurred during this financially-motivated scam.
• Economic security: Minor. Bypassed retail banking security protocols and defrauded individuals of over $35 million, demonstrating vulnerability in consumer financial systems but without causing systemic macroeconomic instability.
• Societal stability: Minor. Caused severe psychological distress and financial ruin for thousands of individual victims, along with localized privacy violations, but did not threaten large-scale social cohesion or civil liberties.