Ingress Component
The Ingress component is the entry point for user inputs to the LLM system. It handles user queries before they are processed by the LLM service.
Component Overview
The Ingress component receives and pre-processes user inputs before they are sent to the LLM service for processing. It serves as the first line of defense against malicious inputs and is critical for ensuring system security.
Security vulnerabilities in the Ingress component can lead to various attacks including prompt injection, unauthorized access, and service degradation through excessive consumption of resources.
Related Vulnerabilities
Prompt Injection
Manipulating LLMs via crafted inputs can lead to unauthorized access, data breaches, and compromised decision-making.
Improper Output Handling
Improper Output Handling refers specifically to insufficient validation, sanitization, and handling of the outputs generated by large language models before they are passed downstream to other components and systems.
Unbounded Consumption
Unbounded Consumption occurs when a Large Language Model (LLM) application allows users to conduct excessive and uncontrolled inferences, leading to risks such as denial of service (DoS), economic losses, model theft, and service degradation