Overview: Trusting Eve

Last updated: September 30, 2026

Overview

Hallucination is a real risk with any generative AI system, and in legal work the cost of an unverified fact is higher than in most fields. Eve is built on a trust-but-verify premise: reduce the chance of a wrong answer, and make checking any answer fast enough that checking becomes routine.

This page is the map. Each capability below has its own article with the detail.

How Eve Reduces the Risk

  • Grounding. Responses are built from the documents in the matter, not from general knowledge. A fact that is not in the matter has no source to come from.

  • Automated validation. Eve checks its own answers after generating them, detecting quoted material and testing it against the source documents.

  • Citations on every response. Each part of an answer shows which document it came from, so the path back to the source is one click.

  • Scoped research. Case law comes from a purpose-built database of official court sources rather than model recall, and web research is restricted to sources the firm approves.

How to Verify What Eve Gives You

Where you are

What to use

Read more

In a chat response

Quote verification highlights every quoted passage and shows whether it matches the source.

📄 Quote Verification

Anywhere in a response

Citations show which document supports each section and jump you to it.

📄 Citations

In a draft

Highlight a factual statement and open its sources to see the exact quotes behind it.

📄 How to Verify Information In Drafted Documents

On case law

Citation history flags, confirmed against the underlying citing references.

📄 Case Law Sources, Coverage & Verification in Eve

On any output

The general review workflow.

📄 How to Verify Eve's Responses

What Verification Cannot Catch

Every tool above answers the same question: is this statement supported by the source? None of them answers whether something is missing.

Eve can be accurate about every fact it states and silent about a record nobody uploaded. Completeness stays with the person running the matter. See 📄 AI Hallucination and Eve's Safeguards.

Working Practice

  • Select context deliberately. A focused document set produces a more accurate answer than the whole file does.

  • Verify before the output leaves Eve, not after someone else has relied on it.

  • Treat verification as routine rather than as something reserved for answers that look wrong. The answers that look right are the ones that go out unchecked.

  • Report anything inaccurate. Reported issues are investigated and fed back into the platform.