HalluCase — The Legal AI Hallucination Registry

A living CVE-style registry for tracking, reproducing, and preventing hallucinations in legal AI outputs.

---

The Problem

In 2025, 729+ documented incidents of AI hallucinations in legal contexts were reported. These incidents resulted in:

• $31K+ in fines for citing non-existent cases
• ABA sanctions against attorneys who submitted AI-hallucinated briefs
• Breached contracts containing fabricated clauses
• Erroneous legal opinions based on hallucinated precedents

HalluCase exists to systematically track these incidents so the legal profession can learn from them, build detection tooling, and pressure AI providers to fix the underlying issues.

---

Quick Start

# Start the API server
npx hallucase serve

# Or clone and run from source
git clone https://github.com/tensflare/hallucase.git
cd hallucase
npm install
npm run serve

The API will be available at http://localhost:3457.

---

CLI Usage

# Search the registry
hallucase search --query "fake citation" --severity critical

# Get a specific report
hallucase get HC-000001

# Submit a new report (interactive)
hallucase submit

# Submit from JSON
hallucase submit --file report.json

# View registry statistics
hallucase stats

# Import from external sources (CSV/JSON)
hallucase import data/charlotin-export.csv

---

API Overview

Method	Endpoint	Description
GET	/api/reports	Search/filter reports
GET	/api/reports/:id	Get report by UUID or HC-ID
POST	/api/reports	Create a new report
PUT	/api/reports/:id	Update a report
DELETE	/api/reports/:id	Delete a report
GET	/api/stats	Registry statistics
GET	/api/schema	Schema definition
GET	/health	Health check

---

Report Schema

Each report receives a unique identifier in the format:

HC-XXXXXX — where XXXXXX is a zero-padded sequential number.

Core Fields

Field	Type	Required	Description
id	UUID	Yes	Unique identifier
hc_id	HC-XXXXXX	Yes	Human-readable ID
title	string	Yes	Concise summary (5-200 chars)
description	string	Yes	Detailed description (min 20 chars)
hallucination_type	enum	Yes	Category of hallucination
severity	enum	Yes	Impact level
domain	string	Yes	Legal domain
hallucinated_output	string	Yes	What the AI produced
expected_correct_output	string	Yes	What should have been produced

Hallucination Types

• fake_citation — Non-existent case or document citations
• misquoted_statute — Incorrectly cited statutes or regulations
• fabricated_contract_clause — Made-up contract provisions
• incorrect_legal_standard — Wrong legal test or standard
• fabricated_obligation — Non-existent legal obligations
• misstated_jurisdiction_law — Incorrect statements about jurisdiction
• incorrect_procedural_rule — Wrong court procedures
• hallucinated_precedent — Fabricated case law precedents
• other — Other types of hallucinations

Severity Levels

Level	Meaning
critical	Would cause certain sanctions or adverse outcome
high	Likely to cause material harm
medium	Potentially misleading but less severe
low	Minor inaccuracies
info	Informational, no direct harm

---

How to Contribute

We welcome contributions! Please see CONTRIBUTING.md for details.

1. Fork the repository
2. Create a feature branch
3. Submit a Pull Request

---

License

Apache 2.0 — see LICENSE.

Built by Tensflare. Open infrastructure for trustworthy legal AI.