Pindrop Takes on AI Voice Deepfakes: Why Your Contact Center Is Already Under Attack

Sumant Mauskar explains how AI-powered voice fraud is attacking contact centers at scale, why the NICE integration matters, and what bi-directional trust really means

Contact centers have an AI problem, and asking agents if calls "sound weird" isn't going to fix it.

According to Sumant Mauskar, Global Partners SVP at Pindrop, humans can detect video deepfakes around 60% of the time. For audio? That drops to 35%. As synthetic voice quality improves daily, traditional fraud detection methods are failing at exactly the moment fraudsters are scaling up attacks.

"You need good AI to fight bad AI," Mauskar says.

He points to a retail client that discovered AI bots requesting $21 returns across thousands of calls. Each amount sat below agent authorization thresholds, so the money went out without question. Death by a thousand cuts.

Pindrop's recent native integration with NICE CX One addresses this by analyzing calls in real time without adding customer friction. The system processes voice alongside metadata, providing agents with simple red, yellow, or green risk indicators based on multifactor analysis.

"We get a copy of the call, process it, and pass information back to NICE," Mauskar explains. "The customer experience doesn't change at all."

But the bigger challenge is coming. As legitimate AI agents start calling contact centers on behalf of customers, CX leaders will need to distinguish between good synthetic voices and bad ones. Mauskar calls this bi-directional trust, and believes the window to prepare is closing fast.

For contact centers still relying on knowledge-based authentication and human intuition, the data tells a different story. The fraudsters are already inside.