Introducing Justinian
A new frontier for legal intelligence
Justinian brings state-of-the-art reasoning into legal work, enabling drafting, analysis, and planning under explicit professional constraints. It is not a chatbot. It is an engine for work that must be reviewed, trusted, and signed.
Legal work, however, is a hostile environment for generic AI. Law demands correctness over plausibility, structure over fluency, explanation over confidence, and accountability over autonomy. A legal professional cannot use output they cannot verify, cite work they cannot trace, or submit documents they cannot defend.
Justinian was built to close this gap: to bring frontier-level AI capability into legal practice without compromising professional standards. It is the proprietary AI engine powering HAQQ Legal AI and HAQQ eFirm - not a single monolithic model, but an engineered system optimized for legal reasoning, drafting, and decision support.
The name Justinian honors Emperor Justinian I of the Eastern Roman Empire - the architect of history's most influential legal codification. Where Emperor Justinian unified Roman law for an empire, our Justinian engine unifies legal intelligence for the practitioner.
“The empire's legal system needed repair. There existed three codices of imperial laws and other individual laws, many of which conflicted or were out of date.” - On Emperor Justinian's mandate, 527 AD
Watch Justinian think
Justinian is not a single model answering in one pass. It is an agentic engine that plans, grounds every claim in real law, checks its own work, and defers to a human before anything is signed.
Reads the question
Parses the request, pins the legal question, and identifies every jurisdiction in play before touching a single document.
Everything behind a single answer
The reasoning above runs on a full stack - product surfaces, security and governance, frontier models, grounded data, and coverage across 30+ jurisdictions.
Samples & Use Cases
Justinian supports the full spectrum of legal work: drafting, research, risk analysis, litigation support, due diligence, and more. Experience it in action.
Your prompt
“Under UAE law, can a shareholder be forced to sell their stake in a private joint-stock company?”
Justinian response
Forced Share Transfer — UAE
UAEUnder Federal Decree-Law No. 32 of 2021 on Commercial Companies, a shareholder cannot be compelled to transfer shares unless the Articles of Association contain a valid drag-along or squeeze-out clause agreed in advance.
Absent such a clause, a forced transfer requires either a court order or the shareholder's written consent. A majority resolution alone does not override this protection.
Justinian flags the two AoA clauses that would need review and links the governing article inline — every statutory reference is grounded in source, never asserted from memory.
Every Justinian output is grounded in source and built for review by a licensed lawyer before it is relied on.
How we evaluate the engine
We hold Justinian to the same standard as our public Legal AI Index: every figure is labelled by how we know it, and every legal claim traces to a source. Then we measure it - against frontier models, and on real end-to-end legal work.
Frontier-Engine Reasoning
Justinian is developed using frontier-engine reasoning techniques combined with legal-specific evaluation benchmarks. The training process emphasizes clause-level validation, jurisdictional accuracy, and continuous expert feedback from practicing legal professionals across multiple jurisdictions including UAE, KSA, Qatar, and beyond.
Measurable Impact
Changes are accepted or rejected based on measurable impact on legal correctness, coverage, and review time reduction. We prioritize improvements that demonstrably reduce the time lawyers spend on routine tasks while maintaining or improving output quality as measured by expert review from licensed practitioners.
Real-World Evaluation
The engine undergoes regular evaluation against internal benchmarks derived from real legal workloads, ensuring that performance metrics reflect practical utility rather than abstract linguistic measures. This evaluation framework includes depth of legal analysis, clause-level coverage, risk identification accuracy, and client-ready formatting compliance.
Every claim carries its provenance
The standard that governs the HAQQ Legal AI Index governs Justinian's own evaluation. Every figure is labelled by how we know it, and every legal claim it produces is traced to a citable source.
Pre-processed from HAQQ's real usage data or our own measured cross-model benchmark.
Third-party forecasts and publicly documented facts, linked to their source.
Directional scenario estimates, always shown behind an illustrative badge - never presented as measured.
HAQQ Legal Agent Study
Long-horizon evaluation built to measure whether agents can do real legal work end-to-end - not just answer trivia. 1,372 tasks, 24 practice areas, ~78,000 expert rubric criteria. Civil-law and MENA coverage by design, all-pass grading by default.
Long-horizon legal agent capability over time
% of Study-equivalent tasks passing all-rubricSame curve shape Karpathy described for coding agents - flat for ~24 months, then a sharp turn around Q1 25 once tool-use and long-horizon planning matured. Legal hit it later because evaluation infrastructure didn't exist.
Legal Agent Study v1 — All-pass leaderboard
% of long-horizon tasks where every rubric criterion passes (best of 3)
Light grey bar = same model's prior-generation baseline (12-month look-back). Best of 3 runs per task on synthetic + anonymized matters.
Mirroring Real Legal Work
All-Pass Grading
M&A change-of-control rubricA deal-team report that catches 8 of 10 risks is not 80% useful. The two missed could change deal economics or surface post-close.
Rubric Anatomy - 57 atomic criteria
M&A change-of-control task24 practice areas — 1,469 tasks
Click a category to filterLimitations
Despite the leap in capabilities, the engine exhibits several limitations common to legal generation engines. Understanding these boundaries is essential for responsible use.
Input Dependency
Incomplete or incorrect facts produce incomplete or incorrect analysis. The engine cannot independently verify factual claims - quality in determines quality out.
Knowledge Boundaries
Cannot access information outside training data and connected sources. Real-time legal updates require explicit integration with your firm's data.
Professional Review Required
Every output requires review by a qualified professional. This is not a limitation to be solved but a design principle.
These limitations are particularly important in our work on legal reasoning engines, which need to accurately represent jurisdictional requirements and professional obligations. We are actively researching ways to address capability limitations without compromising accountability.
Legal responsibility always remains with the lawyer. This is intentional.
Safety & Professional Responsibility
Justinian is built around four non-negotiable obligations that govern every interaction. The engine will never act autonomously, never replace legal judgment, and all actions remain reviewable by a human professional. Learn more about our enterprise security practices.
Disclosure
AI-generated content is always labeled. Users know what comes from Justinian and what comes from human work. Full transparency in every document.
Competence
Operates within trained legal domains. Refuses to speculate outside its knowledge boundaries. Domain-restricted for professional accuracy.
Confidentiality
Zero data retention policy. End-to-end encryption with AES-256. No training on user data.
Oversight
Every output requires professional review before client delivery or filing. Built for enterprise workflows with human-in-the-loop checkpoints.
Built for Compliance
Justinian's architecture aligns with GDPR, ISO 27001, SOC 2 Type II requirements. For in-house counsel and law firms handling sensitive matters, our security infrastructure ensures your data never leaves your control.
AI Research Behind Justinian
Deep dives into the architecture, experiments, and benchmarks that power our legal AI engine.
- 01Research12 min
AI Document Review Software in 2026: Beyond RAG and Chatbots
Read article - 02Research25 min
Legal Ontology AI: How We Cut Legal AI Costs by 97%
Read article - 03Research26 min
Legal AI Predictions to 2030: What 72 AI Agents Forecast
Read article - 04Research8 min
NotebookLM for Lawyers: Memory With Search, Not Legal AI
Read article - 05Research12 min
We Benchmarked 7 LLMs on New York Litigation Strategy
Read article - 06Research12 min
Human-in-the-Loop AI: The Definitive Guide for Lawyers (2026)
Read article
The measured research behind Justinian
Cross-model legal benchmark
How Justinian scores against frontier models on legal tasks.
View the graphAI autonomy scores
How much independent action legal AI should take - and where it must stop.
View the graphHallucination rates
Why grounding every claim in source is non-negotiable for legal work.
View the graphCross-Model Benchmark
HAQQ's measured benchmark scoring legal AI models out of 50.
Legal AI Index
The full data report on how AI is used across the legal ecosystem.
Legal Agent Study
All-pass rubric grading of AI agents on long-horizon legal work.
Research Hub
Browse every legal AI research graph in one place.
Frequently Asked Questions
How is Justinian different from ChatGPT?+
ChatGPT is a general-purpose language model that generates plausible-sounding text. Justinian is a proprietary legal AI engine built specifically for legal reasoning - it applies legal rules to facts, respects jurisdictional boundaries, cites real sources, and produces structured, client-ready deliverables. It doesn't hallucinate laws or cases.
Does Justinian cite real case law?+
Yes. Justinian searches verified legal databases and authoritative sources before answering. Every citation is traceable and verifiable. When sources conflict or the law is ambiguous, Justinian flags the uncertainty rather than presenting a confident but wrong answer.
Can Justinian draft contracts in Arabic?+
Yes. Justinian drafts natively in Arabic, English, French, and other languages with full RTL support. It understands legal nuance in each language and can produce bilingual contracts while maintaining legal precision across both versions.
What jurisdictions does Justinian support?+
Justinian supports 80+ countries including the UAE, Saudi Arabia, Lebanon, Egypt, US (federal and state), UK, EU, France, and international arbitration frameworks like ICC and UNCITRAL. Coverage expands continuously.
Is Justinian's reasoning auditable?+
Yes. Every Justinian output includes a transparent reasoning chain - you can see which rules were applied, which sources were consulted, and how the conclusion was reached. This auditability is critical for professional accountability and client trust.
