IVRy Acquires Patent for Proprietary Voice AI Architecture Preventing Hallucinations

IVRy Inc. (Headquarters: Minato-ku, Tokyo; Representative Director/CEO: Ryoga Okunishi), the developer and provider of the conversational AI platform 'IVRy', is pleased to announce the acquisition of a patent for a proprietary architecture that separates the generation of 'clarifying questions' (a function where the AI understands the context and spontaneously asks questions) in the voice AI domain.

This patent is registered under the H04M category (the technical field relating to 'telephonic communication' in the International Patent Classification). It is a unique conversational AI technology that automatically generates clarifying statements based on context understanding while simultaneously suppressing hallucinations and maintaining natural voice interactions.

The patent overview is as follows: Patent Number: No. 7865495; Patent Holder: IVRy Inc.; Title of Invention: Voice automated response apparatus, voice automated response method, and voice automated response program; Registration Date: May 18, 2026.

Regarding the background, recent years have seen a full-scale adoption of AI agents in customer service frontlines, such as retail stores and contact centers. However, in actual operations, instances where customer utterances are ambiguous or contain multiple intents occur frequently. In such scenarios, it is essential for the AI to spontaneously ask appropriate questions before providing a definitive answer to accurately narrow down the customer's intent. Conventional approaches failed to automatically generate optimal questions tailored to situations outside pre-defined scenarios, leading to issues where intents could not be fully identified, resulting in incorrect answers or broken conversations. While IVRy has provided highly reliable conversational AI services by combining multiple AIs to suppress hallucinations, there has been an increasing need for more flexible voice interactions. Generally, in generative AI technology, balancing the elimination of hallucination risks with ad-hoc conversational flexibility according to the situation has been considered extremely difficult.

The newly acquired patent employs a unique separated architecture. It does not let the AI generate the final answers to the customer; instead, it only allows the AI to automatically generate the wording of the 'clarifying questions' needed to narrow down the customer's intent. This makes it possible to identify intents faster and reach resolutions efficiently.

The core features include: First, highly accurate intent recognition from ambiguous customer utterances, expanding the scope of automated intent identification and reducing operator transfer costs. Second, final answers are selected exclusively from standardized data where there is zero room for hallucinations, providing peace of mind that the AI will not lie, lowering barriers to entry in critical sectors like contact centers. Third, AI generates real-time additional questions only when the intent is ambiguous, allowing continuous dialogue without breaking, while reducing the burden of exhaustive scenario design.

The superiority of this patented technology lies in its 'Role Separation Architecture', which completely separates the 'First Machine Learning Model' for intent identification and answer selection from the 'Second Machine Learning Model' for generating context-aware clarifying questions. This novelty and inventive step were officially recognized by the Japan Patent Office.

CEO Ryoga Okunishi commented that as voice AI agent adoption accelerates in 2026, he is confident this technology will serve as a robust foundation for enterprises to confidently utilize advanced systems, continuing their mission to deliver the best technology to everyone.