4TB of Voice Data Stolen from 40,000 Mercor AI Contractors

On April 4, 2026, the hacking group Lapsus$ released a massive 4TB data archive from Mercor, a broker for AI training data, on its leak platform. This data includes voice recordings, government IDs, and selfies from more than 40,000 contractors who participated in reading and verification tasks. The breach stands out because it combines clear voice samples with authenticated identity documents, paving the way for synthetic voice cloning. According to The Wall Street Journal in February 2026, about 15 seconds of clear audio is needed for high-quality voice cloning, while Mercor's recordings typically last 2–5 minutes. These cloned voices could be exploited for fraud, including bypassing bank security, scamming, and deepfake calls. In response, five lawsuits were filed by contractors within ten days, claiming their voice data was collected under misleading pretenses. ORAVYS is providing free forensic analysis for affected individuals, utilizing advanced detection technologies.