The Wall Street Journal reported in February 2026 that high-quality voice cloning now requires roughly fifteen seconds of clean reference audio for tools available off the shelf.
15秒的干净参考音频是高质量语音克隆的门槛,而Mercor泄露的数据平均每个承包商有2-5分钟的录音,远超过这一阈值。这意味着攻击者可以使用泄露的数据创建非常逼真的语音克隆,大大增加了数据被滥用的风险。