AI Hallucination and Eve's Safeguards

Last updated: March 10, 2026

Overview

AI hallucination is when an AI system generates information that appears plausible but is fabricated or inaccurate. This risk exists across all AI platforms, including Eve, and can affect both legal and non-legal facts. Eve has implemented two built-in safeguards — Fact Search and Quote Verification — to help users identify and mitigate hallucinated information.

How It Works

Why Hallucination Occurs

Hallucination happens when the underlying Large Language Models (LLMs) use general information they've learned during training to inform current answers, rather than relying solely on the specific documents or context provided by the user. This can result in the AI "making up" details that sound correct but aren't supported by the actual source material.

Eve's Two Safeguards

Eve uses two features to reduce the risk of hallucination:

  1. Fact Search — An exhaustive search tool that reads through large numbers of documents page by page and generates a table of key facts and quotes related to a specific query. Eve interprets the user's request, processes documents systematically, and runs the entire output through quote verification to ensure accuracy.

  2. Quote Verification — An automatic validation system that runs on any quote Eve provides in a response. Eve performs a series of checks to verify that each quote exists in the provided documents, helping users quickly assess whether cited information is trustworthy.

Key Terms

  • AI Hallucination: When an AI system fabricates information that appears plausible but isn't supported by source documents or factual records.

  • Fact Search: Eve's feature for performing exhaustive, page-by-page searches across documents to identify all evidence related to a query, with results presented in a verified table format.

  • Quote Verification: Eve's automatic validation system that checks whether quoted text from Eve's responses actually exists in source documents, displayed through color-coded highlights (green for exact match, yellow for partial match, red for not found).

  • Verification States: The three color-coded indicators (green, yellow, red) that show whether Eve successfully found a quote in the source documents.