Agentic AI: Pindrop Anonybit

Lisa Ernst · 26.10.2025 · Technology · 5 min

The increasing spread of agentic AI, i.e., AI agents that autonomously pursue goals and carry out actions, poses new challenges for fraud detection and identity verification. Two companies, Pindrop and Anonybit, offer different approaches to address these threats while making the potentials of autonomous agents safely usable.

Source: Custom representation

Introduction

Agentic AI refers to AI agents that pursue goals, plan steps, and autonomously perform actions. This can include making calls, joining meetings, or initiating transactions. This development leads to an increased threat from machine-led voice fraud and deepfakes.

Pindrop focuses on real-time detection of deepfakes and anomalies in calls and meetings. Their solution Pulse for Meetings analyzes audio and video in web conferences to identify fraud attempts. Anonybit, by contrast, addresses the identity binding of AI agents through a decentralized biometrics infrastructure. Biometric templates are distributed into Shards across multiple nodes to avoid a single point of failure while keeping the identity verifiable. The idea of identity-bound agents cryptographically links AI actions to real people who must authorize them.

Technologies and Actors

Pindrop specializes in voice security and offers real-time deepfake checks in web conferences, including Webex. The solution Pulse for Meetings analyzes audio and video to detect machine-led voice fraud. Pindrop warns of a growing threat from agentic AI , which could significantly increase deepfake fraud in 2025.

Anonybit is developing a decentralized biometrics infrastructure. Biometric templates are distributed in Shards across multiple nodes , to prevent a single point of failure. This enables verifiable identity without central biometrics silos. The concept of the identity-bound agents ensures that AI actions are cryptographically bound to real people and are only executed after approval by a living, authorized person. This is ideally done multimodally (voice, face, possibly iris) and privacy-preserving.

Current Developments

On April 28, 2025, Pindrop announced the beta of Pulse for Meetings a real-time detection of audio- and video-deepfakes directly in Zoom, Teams and Webex. On October 20, 2025 the expansion followed: Pindrop integrated detection and passive voice biometrics into the Webex Suite and the Webex Contact Center. Meetings receive multimodal fake checks, Contact Center a user-friendly authentication. Pindrop repeatedly locates the threat landscape in guides and webinars, including a projected substantial increase of deepfake fraud in 2025.

Anonybit launched on May 28, 2025 secure agentic workflows , which, according to their own statements, is the first productive implementation of agentic commerce scenarios with decentralized biometrics. This was followed by integration paths and partnerships, including an identity layer for AI agents with the no-code platform SmartUp on July 7, 2025. Technically, Anonybit relies on patented methods for decentralized, multimodal biometrics (voice, face, fingers, iris).

Analysis and Context

Pindrop's motivation lies in the rising danger that agents and attackers hijack meetings with convincingly real voices and avatars, manipulate decisions, or disrupt contact-center workflows. The strategy is deep integration into large collaboration platforms such as Webex to minimize friction and alert security teams in real time. The Integration with Webex is an example of that.

Anonybit addresses the problem that agents without robust, privacy-preserving identity binding ultimately remain unassignable to anyone. Decentralized storage and cryptography (e.g., MPC/ZKP) enable verification without central data silos, which is useful for compliance and liability chains. Research shows deepfake detection remains challenging and accuracy varies depending on dataset and attack. This argues for a layered defense.

Evidence shows that Pindrop offers multimodal deepfake detection for meetings and is integrated into Webex, complemented by passive voice biometrics and device intelligence in the contact center. It is also evidenced that Anonybit launched secure agentic workflows and a decentralized biometrics cloud, including a patent for decentralized biometric authentication.

Concrete, independent metrics for false positive/false negative rates of meeting detection in heterogeneous real-world scenarios are unclear. The literature cautions that detection performance depends strongly on the dataset and attack type. It is also unclear how interoperable identity-bound agents are across platform boundaries and how governance and revocation are practically solved. The assumption that decentralized automatically means GDPR-secure is misleading, as choosing off-chain approaches can strengthen privacy but does not replace proper data classification, purpose limitation, and access control. The assumption that biometrics solve all privacy risks is also misleading, as compromised biometric traits cannot be easily reset.