Multi-LLM Group Chat Consent Control for Secure Agent Invocation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing AI systems with multiple large language model (LLM) chatbots face challenges in ensuring security, privacy, and data use restrictions due to lateral data sharing, non-deterministic behavior, and lack of user control over data access and communication between agents, which can compromise sensitive information and copyrighted content.

Innovation Solution

Implementing a security module that handles customer authentication, authorization, and in-context customer consent to control which LLM agents and components can access chat message history, with features like private mode and user-controlled data sharing, using industry standards like OpenID Connect and OAuth2.0, and applying security, privacy, and data use restrictions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If multiple LLM agents are allowed to communicate and share data laterally to improve system capabilities, then the system's adaptability and versatility are improved, but security and privacy are compromised due to unauthorized data sharing

Engineering Contradiction:
Improvesystem capabilitiesVSAvoidsecurity and privacy
Core Design Contradiction:
Adaptability or versatilityVSObject-affected harmful factors

Solution Approach 1:

The patent introduces a security module as an intermediary between LLM agents that mediates all data sharing and communication. This module evaluates data requests, checks authorization policies, and controls lateral data flow between agents, enabling versatile agent collaboration while maintaining security through centralized oversight and policy enforcement.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Adaptability or versatility

If LLM agents operate autonomously with non-deterministic behavior to improve system intelligence, then the system's adaptability is improved, but predictability and control are reduced making security guarantees harder to achieve

Engineering Contradiction:
Improvesystem intelligenceVSAvoidpredictability and control
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The security module implements continuous feedback mechanisms by monitoring LLM agent operations, evaluating data sharing requests in real-time, and enforcing authorization policies. This feedback loop provides deterministic control over non-deterministic agent behavior, ensuring security guarantees while preserving agent intelligence and adaptability.

Inventive Principle:
Principle #23Feedback

3Object-affected harmful factors

If user control mechanisms are implemented to improve security and privacy, then security is improved, but system complexity increases due to additional authentication and consent management layers

Engineering Contradiction:
Improvesecurity and privacyVSAvoidsystem complexity
Core Design Contradiction:
Object-affected harmful factorsVSDevice complexity

Solution Approach 1:

The security module serves multiple functions simultaneously: authentication, authorization, data use restriction enforcement, and consent management. By consolidating these security-related functions into a single multi-functional module, the system achieves comprehensive security without proportionally increasing complexity, as the module handles diverse security requirements through unified policy-based control.

Inventive Principle:
Principle #6Universality (Multi-functionality)

4Object-affected harmful factors

If consent requests are required for each agent invocation to improve user control and security, then security and user control are improved, but operational efficiency and response time are reduced

Engineering Contradiction:
Improveuser controlVSAvoidoperational efficiency
Core Design Contradiction:
Object-affected harmful factorsVSProductivity

Solution Approach 1:

The system performs preliminary authorization by establishing user preferences and authorization policies before agent interactions occur. The security module pre-evaluates potential data sharing scenarios based on stored user preferences, allowing rapid consent decisions without requiring detailed real-time user input for each agent invocation, thus maintaining user control while improving operational efficiency.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4657277A1Enabling consent in a group communication with multiple LLM agents
Publication Date: 2025.12.03 MICROSOFT TECHNOLOGY LICENSING LLC
  • EP4657277A1 patent drawingFigure 1
  • EP4657277A1 patent drawingFigure 2
  • EP4657277A1 patent drawingFigure 3

AI summary

Security, privacy, and data use restrictions in group communication with multiple large language model (LLM) chatbots or agents are provided. In-context user consent is obtained for operations performed by LLM agents on behalf of the user as and when needed. A first message directed to a first LLM agent is received via a user interface (UI). Based on a determination that the first message is to invoke a second LLM agent, a consent request for consent of a user to invoke the second LLM agent is provided via the UI. Upon receiving the consent of the user to invoke the second LLM agent, the second LLM agent is invoked within the context of the UI. In some examples, a command to enter private mode may be received to limit the communication in the private mode only selected LLM agents.