How Agentic AI, Pindrop And Anonybit are killing Voice Frauds In 2026

This example occurred: In 2019, a fraudster called the UK-based energy company’s CEO. Same voice, same accent, same urgency, same cadence, it was his boss. Before he realized that it was not a human voice, the CEO had wired $243,000. That attack utilized voice cloning technologies that were in their infancy. By 2026 things are looking very different, tools are more accessible, more scalable, and much more convincing. A sophisticated fraud ring used to be what was needed, but now it’s three seconds of stolen audio and a free text-to-speech engine.

That’s the world that agentic AI Pindrop Anonybit was created to solve. All three of these technologies are combined to create a multi-layered protection against not just scammers, but AI-powered attackers that are able to impersonate, infiltrate and defraud at a rapid rate. It is imperative that anyone in a banking, healthcare, insurance or enterprise security role has a solid grasp of the reason for each layer and what they do with one another.

Securing the world means securing yourself; and the world changes, so do security measures.

The passwords, security questions and IP blocking are created to protect against single users with static tools. It’s no longer this world. Modern fraud rings leverage generative AI to initiate thousands of credential-stuffing attacks at once, switching out proxies that match real traffic, and voice clones created using audio content from social media. A static defense is one that does not continually adapt, reason and change; one that does not adapt on the first time an opponent learns the rules of that defense.

Agentic AI is a transformative decision level technology. 

An agentic system is a system that continually works towards an objective, whereas a traditional AI system waits for the user to instruct it. In terms of identity security, it means monitoring sessions across a range of different channels, connecting together behavioural signals, real-time scoring risk, and perhaps redirecting or preventing a risky interaction before a human analyst is even notified. Research has shown that, in all the agentic security systems deployed, autonomous threat response has cut incident response time by more than 50%. That’s where that speed advantage comes in, as a fraud attack occurs in milliseconds, a stopped transaction is different from a cleared transaction.

Voice fraud is no longer about the future, it is here, and it is growing faster than anyone could have imagined. Deepfakes, synthetic voice technology and AI-powered scams are changing the face of fraud.  Vijay Balasubramaniyan, CEO & Co-Founder, Pindrop

The agentic AI fraud problem is exacerbated by the fact that fraudsters are also leveraging autonomous tools, like the defenders. By using an AI agent, the fraudster can dial a bank’s IVR, follow the prompts, record the voice pattern of a trusted customer and use it to request a wire transfer, all without raising a finger. Security systems for legacy were created for human use. Given this new reality, the agentic AI Pindrop Anonybit stack is made for it.

Catching what human Ears Miss 

Enterprise voice security startup Pindrop unveils ‘the voice intelligence layer’. More than five billion calls have been already analyzed by Pindrop since it was launched. The company is located in Atlanta, GA, USA.  The technology is located at the level of the contact center and processes more than 1,300 acoustic, behavioral and device-level features simultaneously and in milliseconds. This output is a risk score that indicates to the system if the voice heard is a real human voice or a synthetic voice.

It’s not what people say but the physics of how they sound that what Pindrop detects. The micro-traces created by synthetic voices generated by text-to-speech (TTS) engines are not detected by human ears: unnatural frequency distributions, lack of variation in breath-pauses due to speech, abnormal resonance profiles and subtle compression artifacts from digital to analog conversion. 

Pindrop’s Pulse product identifies these patterns in about two seconds before a contact center agent can utter a single word. It was recently deployed in early 2026 for healthcare and claimed a 99.2% accuracy rate for the combination of liveness detection and voice authentication, while assisting to limit a coordinated attack against 1,200 patient accounts resulting in an estimated $18 million in potential fraud exposure.

Pindrop natively integrates to the platforms that most large contact centers already operate: Amazon Connect, Genesys, Cisco Webex, and Five9. Organizations are not required to build out their infrastructure, they add Pindrop’s voice intelligence to what they already have. According to Pindrop’s 2025 Voice Intelligence & Security Report, contact centers experienced 2.6 million fraud events in 2024, resulting in losses of $12.5 billion, and the total could rise to as much as $44.5 billion in 2025 without scaling detection tools. If Pindrop detects a call as high risk, it conveys that score to the agentic decision layer, which then decides what to do next.

Anonybit: Solving the Biometric Honeypot Problem for Good.

Every organization holding a central repository of biometric data is on a so-called honeypot – a single, high-value target – that security practitioners rely on as a term for a central target with the potential to house a large number of valuable data types. But if that database is hacked, then all the identities enrolled are compromised forever. You can issue a new password. A new fingerprint, face or voiceprint cannot be issued. Frances Zelazny, one of the founders of Anonybit, created the company in 2018 to solve this issue in architecture.

Rather than keeping a full biometric record in any place, Anonybit’s platform breaks each biometric template into shards of information that are encrypted using techniques inspired by multi-party computation (MPC) and zero-knowledge proofs (ZKP). These shards are then spread out in many decentralized cloud nodes. Even when one node is compromised, the attacker will get meaningless encrypted noise if they want to recreate an easy-to-use identity.Verification still works: the system does not need to reconstruct a complete record in order to verify, it verifies without needing a complete record. Anonybit calls this the Circle of Identity. 

Agentic commerce has incredible potential in terms of scale and efficiency, but also great risks of trust, fraud, and control without identity. Our decentralized biometric cloud, data vault and token management system is the missing identity layer required by the enterprise. Frances Zelazny, Co-Founder & CEO, Anonybit

A real world scenario of how the Three layers work together.

Agentic AI Pindrop Anonybit is not about any individual tool, but the way these tools interact when it matters. Think of a fraud attack that security teams reported in 2025, in which a fraud ring was able to acquire 45 seconds of a banking customer’s voice from a public earnings call recording. They fed it into a commercial TTS application; duplicated the voice; and developed an agentic bot to call the contact center of the bank and request a wire transfer authorization.

The reaction of each layer is given below. Pindrop’s Pulse was able to detect the TTS artifacts within less than 400ms after the incoming audio was received, and return a risk score of 87 out of 100. In addition, Anonybit was able to do a biometric shard verification service for the claimed account, and they could not find a match on their end.

Both signals were sent to the agentic AI, which correlated them with an unknown device fingerprint and ended the session without further action, and marked the account for review. The whole sequence was done in less than 200ms. The next call from the legitimate customer was without any abnormality. The fraud bot was not able to get to a real agent.

With all three layers, organisations can reach up to 80% reduction in fraud and as quickly as 10 seconds per authentication call. In addition, all activities performed by an AI agent need to be cryptographically linked to the authenticated human user who issued the instructions, offering a complete and verifiable chain of custody for regulators and audit teams.

Three Shifts Security Teams need to Watch

The combined use of agentic AI, voice intelligence, and decentralized biometrics is still in its infancy. However, three trends are already influencing its direction of travel. The first is that the detection of deepfakes is moving beyond voice. 

Pindrop’s Pulse for Meetings, which was launched in beta in April 2025 and can be expanded to Webex in October 2025, uses the same acoustic physics analysis for live video conferencing to identify synthetic faces and voice artifacts in real-time Zoom, Teams or Webex meetings. The same fraud issue affecting contact centers is making the rounds in board rooms and negotiating phone calls.

Second, AI agents will have to be endowed with credible credentials of their own. Now, an AI agent performs on behalf of a user. AI agents will engage in negotiations with other AI agents, trigger payments and approve supply chain actions in fully automated pipelines tomorrow.

Third, quantum-resistant cryptography is a design requirement, not a future consideration. Anonybit’s data vault is already decentralized, and that is important because today’s encrypted biometric shards must be secure against computing power that doesn’t yet exist, but may at some point.

Agentic AI, the development of digital identity standards, and the ongoing creativity of fraudsters who will exploit weak points, such as account recovery and help desks, will define the next five years.  The author, Frances Zelazny, is a Co-Founder & CEO at Anonybit (SC Media, October 2025).

Final Verdict

The fraud problem is outpacing the tools that are created to combat it. Passwords fail. Security questions fail. Biometric databases are targets. Today, autonomous AI allows attackers to work at a more massive scale and faster pace than a human fraud team in real time. The agentic AI Pindrop Anonybit stack is not the answer to this with a single product, it is the answer with architecture. 

Pindrop captures your ear’s unheard ability! Anonybit deletes the database that hackers desire to steal. Agentic AI is able to link signals and act quicker than any human review process. Organizations that use all three layers do not suffer from being a victim and are truly proactive defenders. That’s not an option for long, and that’s where the entire identity security industry is headed – it’s the direction in which they’re going regardless of whether they are the leader or follower.

Want to know about AI (Artificial Intelligence) Voice Agents, you can read our detailed case study and follow TheTweaks for more tech updates.