Detect Deepfakesby Resemble AI
Deepfake case study · Audio

Is ElevenLabs safe? What happened after I uploaded my voice samples

An analysis of the security and privacy risks associated with AI voice cloning platforms following a major 2024 deepfake fraud incident

Incident date
Jan 2024
Target
an unnamed bank CEO
Updated Jun 26, 2026 · 1 min read

In 2024, a bank CEO was successfully deceived by fraudsters using voice deepfakes, resulting in a $25 million illicit transfer. This incident underscores the urgent security implications of increasingly accurate AI voice cloning technology and the potential for misuse of digital identity.

What happened

The fraud involved the use of synthetic voice technology to impersonate a target, successfully manipulating a bank CEO into authorizing a significant financial transaction. While the incident highlights the destructive potential of high-fidelity voice synthesis, it has also sparked broader concerns regarding how platforms handle voice data.

Security analysis of platforms like ElevenLabs reveals that while some services implement enterprise-grade protections—such as SOC 2 Type II certification, voice verification systems, and celebrity-blocking algorithms—users face significant privacy trade-offs. Standard data handling practices often include the retention of raw voice recordings for up to three years, and more critically, the permanent integration of voice characteristics into AI models. Even when a user deletes their account, these models may persist, effectively creating a permanent digital fingerprint of the user's voice.

Furthermore, platforms utilize content moderation, which can involve human review of inputs to detect policy violations, potentially exposing sensitive data to third parties. For individuals, the risk of data permanence remains a primary concern; once a voice is processed into a model, it may be impossible to fully remove. Security experts suggest that users treat voice uploads as permanent assets rather than temporary data, advising against the use of personally identifiable or distinctive speech patterns to mitigate the risk of long-term identity exposure. As the industry matures, the gap between legitimate enterprise security standards and the privacy reality for individual users remains a significant point of concern for digital safety.

Sources