CyberCertPrepAI Security FundamentalsGenAI Misuse, Deepfakes & AI-enabled AttacksEASYFree question
Which technical property of current voice-cloning tools makes them dangerous even when only a 10-second audio sample of the target is publicly available?
- A.Modern voice-synthesis models can learn a speaker's tonal, pitch, and cadence characteristics from a very small sample and generate arbitrary speech in that voice
- B.Voice cloning requires at least 60 minutes of high-quality audio, so 10 seconds is insufficient for a convincing clone
- C.Voice cloning tools require the target's consent and a signed API agreement before generating audio
- D.Voice characteristics are protected by copyright law, preventing commercial use of cloned voices
Why A is correct
State-of-the-art few-shot text-to-speech models (e.g., ElevenLabs, Tortoise TTS) can produce convincing voice clones from as little as 3-30 seconds of reference audio by capturing speaker embeddings. This makes any publicly posted audio - social media clips, YouTube interviews, voicemail greetings - an attack surface. The 60-minute claim is outdated; consent requirements are policy goals, not technical barriers; copyright law does not prevent real-time generation in this attack scenario.
Know someone studying for AI Security Fundamentals? Send them this one.