This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
In spring 2023, a Florida investor was targeted by a voice deepfake scam where an attacker used AI-generated audio to impersonate him during a call to his bank. The banker recognized the voice as synthetic and repetitive, leading to a failed transfer attempt and the flagging of the account. While the fraud was unsuccessful, the incident highlights the growing risk of AI-enabled voice impersonation in financial services.
In spring 2023, Florida investor Clive Kabatznik became the target of an advanced scam attempt involving a voice deepfake mimicking his own voice. The fraudulent caller, using AI-generated speech, contacted Kabatznik's Bank of America representative in an unsuccessful attempt to deceive the banker into transferring funds to a different account.
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 voice cloning to impersonate a specific individual in order to execute a targeted financial scam against a bank.
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 caused by human scammers intentionally using a post-deployment AI voice cloning tool to attempt financial fraud.
Added categories peak at 2: Minor (Psychological Harm, Direct)
An unknown attacker used generative AI voice cloning to impersonate a Florida investor in a failed attempt to execute a fraudulent bank transfer. While the scam was unsuccessful, it demonstrates the evolving capability of malicious actors using AI for targeted financial fraud, carrying minor national security implications primarily around economic security and synthetic media risks.
The AI system (VALL-E) was used as a tool by human scammers to perform a specific function: generating synthetic speech that mimics a target's voice based on direct inputs (audio samples and text prompts).
Limited Risk: The system is used to generate AI-generated content such as deepfakes, which falls under Risk Level 3 due to transparency obligations to ensure users are aware they are interacting with AI-generated content.
The primary goal of the Microsoft VALL-E system is voice generation, specifically cloning a target's voice from a short audio sample to generate arbitrary speech.
The report provides clear, firsthand details of the specific scam attempt targeting the investor, including the banker's reaction, the subsequent communication lockout, and the source of the audio data used. The role of AI voice cloning is explicitly stated and well-documented.
None
Technology
Generative AI Voice Cloning
AI programs clone voices using very short audio samples.
Low-Cost Generative AI Tools
Cheap tools make sophisticated voice synthesis widely accessible.
Text-to-Speech Software
Type-to-text programs let attackers type what the synthetic voice says.
Data Inputs
Public Voice Recordings
Speeches and social media videos provide ample audio for cloning.
Stolen Customer Data
Stolen bank details sold on underground markets help target victims.
Human Factors
Susceptibility to Voice Spoofing
Employees can be tricked by realistic cloned voices of known clients.
Process and Methods
Caller ID Spoofing
Attackers easily spoof target numbers to bypass caller ID trust.
Inadequate Direct Verification
Lack of robust out-of-band authentication for direct phone calls.
Management
Narrow Security Focus
Securing call centers while leaving direct mobile lines vulnerable.
AI system involved
Identified by automated extraction; can differ from the alleged developers above.
