Voice Authentication Prompts Updated to Block Deepfake Attacks

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Voice-based authentication systems are vulnerable to deep fake attacks, allowing unauthorized access through synthesized voices that mimic legitimate users.

Innovation Solution

A computing platform utilizing a generative AI model generates and stores voice-based authentication prompts on a distributed ledger, scoring user responses to grant or deny access based on a predetermined threshold, and dynamically updates prompts to enhance security.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If voice based authentication is used to verify user identities, then authentication convenience is improved, but security against deep fake attacks deteriorates

Engineering Contradiction:
Improveauthentication convenienceVSAvoidsecurity against deep fake attacks
Core Design Contradiction:
Ease of operationVSReliability

Solution Approach 1:

The system dynamically updates authentication prompts at predetermined intervals and stores them in a distributed ledger. The prompts change over time, making static deep fake recordings obsolete. This dynamic updating mechanism maintains convenience while improving security against replay attacks and deep fakes.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system performs preliminary voice authentication during registration to establish baseline voice characteristics before actual authentication events. This preliminary action creates reference data that enables more accurate detection of deep fake attempts during subsequent authentication events.

Inventive Principle:
Principle #10Preliminary action

2Manufacturing precision

If deep fake generation capability increases, then voice synthesis quality is improved, but authentication security deteriorates

Engineering Contradiction:
Improvevoice synthesis qualityVSAvoidauthentication security
Core Design Contradiction:
Manufacturing precisionVSObject-affected harmful factors

Solution Approach 1:

The system uses the same generative AI technology that creates deep fakes to generate synthetic voice samples for training detection models. By converting the harmful deep fake capability into a beneficial training resource, the system improves its ability to detect such attacks while maintaining voice synthesis quality.

Inventive Principle:
Principle #22Blessing in disguise (Convert harm into benefit)

Solution Approach 2:

The system implements a scoring mechanism that provides feedback on authentication attempts. When deep fake attempts are detected, the system learns from these failures and adjusts its detection thresholds and models, creating a continuous improvement loop that addresses increasing deep fake sophistication.

Inventive Principle:
Principle #23Feedback

3Reliability

If voice authentication prompts are frequently updated, then security against deep fake attacks is improved, but system complexity increases

Engineering Contradiction:
Improvesecurity against deep fake attacksVSAvoidsystem complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The system introduces a distributed ledger as an intermediary to manage and store authentication prompts. This intermediary layer automates the distribution and verification of updated prompts across the network, reducing the manual coordination complexity that would otherwise be required to frequently update authentication challenges.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentUS12499197B2Deep fake attack prevention for voice based authentication leveraging generative artificial intelligence (AI) and distributed ledger technology
Publication Date: 2025.12.16 BANK OF AMERICA CORP
  • US12499197B2 patent drawing
  • US12499197B2 patent drawing
  • US12499197B2 patent drawing

AI summary

A computing platform may generate, using a generative AI model, voice based authentication prompts corresponding to a user. Upon receiving a registration request from the user, the computing platform may identify the voice based authentication prompts for the user. The computing platform may send, to a first computing device of the user, the voice based authentication prompts and may receive/store voice based authentication information corresponding to the voice based audio inputs. Based on receiving an access request, the computing platform may send the plurality of voice based authentication prompts, and may receive, from a second computing device, additional voice based audio inputs. The computing platform may score, based on the voice based authentication information, the additional voice based audio inputs. The computing platform may compare the score to a threshold. Based on identifying that the score fails to meet the threshold, the computing platform may initiate security actions.