The Situation

Contact centers traditionally rely on a combination of knowledge-based authentication and human judgment on the part of agents. Agentic AI attacks exploit that dependency by overwhelming the system with synthetic voices and LLM-driven conversations that are well researched and generally indistinguishable from a human caller. A single fraudster can now orchestrate hundreds or thousands of simultaneous calls operating continuously through autonomous systems running 24/7, overwhelming fraud analyst teams that cannot keep pace with automated attacks at this scale

The characteristics that make agentic AI attacks uniquely difficult to defend against include:

• Scale: automated systems operate across every queue simultaneously, conducting more attack attempts in an hour than a human fraud ring could execute in a month
• Consistency: agentic systems never hesitate, never make the behavioral slips that betray human fraudsters under pressure, and maintain consistent personas across extended multi-step interactions
• Adaptability: LLM-driven agents respond dynamically to unexpected questions, navigate complex verification processes, and adjust their approach when initial attempts fail
• Exhaustive credential deployment: agentic systems can systematically ingest and deploy large volumes of breached PII, testing every credential against every account without fatigue

No human defense scales to meet an automated attack of this volume.

The Solution

Daon’s synthetic voice detection addresses agentic AI attacks at the signal level, which is the only point in the contact center stack where machine-scale automated attacks can be intercepted without human review. Operating from the first seconds of a call, detection identifies AI-generated audio regardless of synthesis model or language, requiring no enrollment and remaining invisible to legitimate callers. Because detection operates at the signal level rather than relying on behavioral cues or conversational inconsistencies, it cannot be defeated by a more convincing script or a better-trained model.

Voice biometric authentication adds a second barrier through its built-in anti-spoofing stack, covering replay detection, synthetic speech detection, pre-seen audio detection, and distorted voice detection. Layering independent detection capabilities means no single point of failure exists for an automated system to exploit.

For organizations employing Identity Continuity, step-up authentication can be triggered at any point in the interaction, requiring the verified account holder to confirm their identity through a face biometric or other strong factor. An automated system has no path through a challenge that requires verification of the person.