The agentic caller always rings at scale: Reality Defender explores new AI voice threat

Deepfakes remain a potent part of the fraud arsenal, but the industry’s alarm bells have begun ringing over agentic AI. A new article from Reality Defender looks at how an influx of AI customer service agents – or “agentic AI callers” – will transform how contact centers work.

Reality Defender VP of Human Engagement Gabe Regan cites numbers from Gartner, which predict that, by 2029, AI will autonomously resolve 80 percent of common customer service issues without human intervention.

“Many of those interactions will originate from AI systems calling in, not humans,” Regan writes. “Contact center infrastructure that engineers built for human callers with occasional bot probes is not prepared for autonomous AI agents at scale, and the gap between what current detection catches and what developers design agentic callers to avoid is where the problem lives.”

His post provides a detailed breakdown of the agentic AI caller problem, which levels up on the robocall by listening to interactive voice response (IVR) prompts, selecting the correct menu options, and conducting “a full adaptive voice conversation with a human agent if the IVR routes it through.” AI generated deepfake audio mimics human cadence and tone, and can adapt to unexpected questions. “Nothing in the interaction pattern flags the call as non-human, because the creators specifically designed the system generating the call to avoid triggering those flags.”

That’s a problem when customer service becomes a purely agentic transaction, in which some agents are calling support centers on behalf of legitimate AI systems. Cracking down on agentic activity risks denying actual customer requests. “Whether the agentic caller is acting on a customer’s behalf or probing for fraud, the contact center needs to know what it is dealing with before the call reaches the queue.”

Regan says the distinction matters operationally. “A human caller calling to resolve a billing dispute is a service request. An agentic AI caller navigating the same IVR and reaching the same agent queue consumes capacity without generating a legitimate human service interaction.” Execute that at scale, and it begins to change the composition of the call queue.

As fraud changes, so must detection strategies

Fraud detection, says Regan, has to change. With agentic AI callers, the behavioral anomalies or unusual activity patterns that give away standard call center fraud do not apply. Agentic AI callers “call from numbers that pass carrier validation, navigate IVR menus at normal speed, select options correctly on the first attempt, and respond to security questions with accurate information. Importantly, they do not fail authentication or generate unusual account activity.”

“Contact center deepfake fraud through agentic callers exploits precisely the behavioral signals that fraud detection relies on, because developers optimize agentic systems to replicate them.”

In short, at scale, agentic callers can warp fraud metrics and efficiency analytics alike through “a gradual change in the composition of inbound volume” – and can cause contact center managers to mistake the two. A rise in average handle times and declines in queue performance look like one kind of problem, when in fact it’s creeping agentic fraud that’s responsible.

Reality Defender offers acoustic detection tools that can identify synthetic voices before they reach the queue, thereby preventing unwanted agents from entering the queue, sapping resources and skewing metrics. Its product analyzes a voice signal for synthetic patterns – compression artifacts, frequency anomalies, and telltale generation signatures – before routing.

Regan says AI bot contact center detection at this layer “works regardless of whether the synthetic caller uses a sophisticated voice clone or off-the-shelf text-to-speech, because the acoustic signal analysis identifies generation artifacts across the full range of synthesis techniques.”

Five years in, Reality Defender is a market shaper

Reality Defender is celebrating recognition from Gartner as a market shaper. In a blog post, Ben Colman, CEO of Reality Defender, reflects on the company’s first five years.

“When we started Reality Defender in 2021, most people still thought of deepfakes as a content problem. A manipulated clip. A social media issue. Something that happened after the fact.”

Generative AI, however, has enabled fraudsters to bombard every possible attack surface in real time. Today, Colman says, “synthetic media reaches contact centers, hiring pipelines, executive communications, investigations, onboarding flows, and enterprise approvals.”

As such, Reality Defender focuses its voice, video, and image detection capabilities across channels enterprises already run, from APIs to contact centers to meetings and investigations.

Colman positions Reality Defender ahead of the curve on AI fraud threats, having built the business on detection exclusively. Detection, he writes, “has to be independent. It has to be multimodal. It has to be fast enough to matter before the decision is made.”

“Deepfake detection is no longer about spotting something strange after the damage is done. It is about giving enterprises a way to verify authenticity before money moves, access is granted, or a candidate advances.”

It’s not what a deepfake is, but how it’s used that matters

Colman’s piece underscores the shift toward continuous monitoring that applies a layered approach – which also features prominently in Gartner’s recent report on how to assess and mitigate risks in identity verification and face biometrics.

Deepfake detection, it says, is highly probabilistic, “with defenders and attackers in a very dynamic battle with one another. There is no standardized approach to deepfake detection, and there is no way to assess the efficacy of what vendors tell you about their particular approach, and there is no way to compare efficacy among vendors.”

In the report’s view, cybersecurity leaders looking for support on deepfakes should focus on
“detecting and preventing the different methods by which a deepfake could be used against a face biometric system.” That means making presentation attack detection a critical requirement, “based on testing against established standards.” Likewise injection attack detection, “seeking evidence of testing against emerging standards.”

“Favor vendors that collect a broad range of contextual signals, such as those related to device and location, to improve their ability to detect attacks.”

Indeed, “cybersecurity leaders should think beyond just ‘deepfake detection’ when it comes to selecting vendors for face biometric use cases such as IDV or authentication. Detecting the actual deepfake itself is not the most effective thing to focus on when attempting to prevent attacks.”

Is deepfake detection being absorbed into reality defense?

The implication is significant for the biometrics market. Deepfake detection as a standalone product matters less than the overall efficacy of a system in detecting fake media across the digital pipeline. A “deepfake attack” is actually an injection attack that makes use of generative AI for video or image manipulation: the deepfake itself is ammunition, but the attack targets hardware or software.

In time, the term “deepfake” may outlive its usefulness. For one, it implies mimicry, when AI can now generate entirely novel synthetic identities. For another, it points to the image or voice signal, rather than the ecosystem that is under attack. Moreover, it tends to be associated with the early examples of generative AI, often applied to celebrities or political figures as a novelty.

We are already far beyond that, and generative AI has been weaponized as a tool for automated fraud at unprecedented scale.

As such, the market for tools to detect AI generated content is likely, over time, to move away from “deepfake detection” in favor of a more holistic offering that simply promises to maintain the fabric of digital reality, regardless of where the threat lies. With that in mind, the name “Reality Defender” looks prophetic indeed.

Article Topics

call centers  |  deepfake detection  |  deepfakes  |  Gartner  |  Reality Defender  |  voice AI