Prompt Processing Units for Direct Prompt Injection Mitigation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing technologies lack the ability to effectively interpret and apply security controls on prompts sent to large language models (LLMs), making them vulnerable to direct prompt injection attacks, which manipulate or expose sensitive information.

Innovation Solution

Implementing prompt processing units (PPUs) to characterize and distill key features from prompts, enabling enterprises to detect and prevent mutually opposed subjects within the prompts, thereby mitigating direct prompt injection threats.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If enterprises use large language models for productivity enhancement, then productivity is improved, but vulnerability to prompt injection attacks increases

Engineering Contradiction:

ImproveproductivityVSAvoidprompt injection attack vulnerability

Core Design Contradiction:

ProductivityVSObject-affected harmful factors

Solution Approach 1:

The patent introduces a prompt processing unit as an intermediary component between the user prompt and the large language model. This PPU analyzes prompts for mutually opposed subjects and prevents injection attacks while allowing legitimate queries to pass through, thus resolving the contradiction between enabling productivity-enhancing LLM usage and preventing security vulnerabilities

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The system performs preliminary analysis of prompts before they are sent to the LLM by identifying mutually opposed subjects in advance. This preliminary detection and prevention mechanism allows the system to maintain both productivity benefits and security against prompt injection attacks

Inventive Principle:

Principle #10Preliminary action

2Adaptability or versatility

If current LLMs interpret and execute prompts autonomously, then task completion capability is improved, but controllability and security oversight deteriorate

Engineering Contradiction:

Improveprompt interpretation and execution capabilityVSAvoidsecurity control and oversight

Core Design Contradiction:

Adaptability or versatilityVSEase of operation

Solution Approach 1:

The prompt processing unit serves as a mediator that sits between the autonomous LLM execution and human security oversight. It maintains the autonomous execution capability while introducing a controllable security layer that can intervene when mutually opposed subjects are detected

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The system segments the prompt processing function into two independent parts: the LLM's autonomous interpretation and execution capability, and the security control layer that identifies mutually opposed subjects. This segmentation allows both high adaptability and maintainable security oversight to coexist

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS20250292018A1Direct prompt injection threat mitigation using prompt processing units

Publication Date: 2025.09.18 CISCO TECHNOLOGY INC

AI summary

In one implementation, a device identifies a first subject indicated by a prompt to a large language model. The device identifies a second subject indicated by the prompt to the large language model. The device determines whether the first subject and the second subject are mutually opposed subjects. The device prevents the large language model from processing the prompt when the first subject and the second subject are mutually opposed subjects.