---
title: "A Tsinghua-born startup turned contract review into a job for Chinese legal AI"
date: 2026-11-12
category: AI in Action
site: NeuroAI
canonical: https://neuroai.site/a/na-app-powerlaw-legal
language: en
---

# A Tsinghua-born startup turned contract review into a job for Chinese legal AI

> PowerLaw, a Beijing company spun out of Tsinghua's computer-science lab, built a legal-specific model that flags risky clauses in enterprise contracts instead of chatting about the law.

A procurement manager at a manufacturing firm opens a 40-page supplier contract. She does not read all of it. She pastes it into a system that, within a minute, highlights the penalty clause buried on page 31, flags a missing liability cap, and suggests a fix with the relevant statute attached. No junior lawyer was billed for the first pass.

That is the everyday use case PowerLaw (幂律智能), a Beijing legal-technology company, has been chasing since 2017. Unlike the general-purpose chatbots that grabbed headlines, its product does something narrower and harder to fake: it reads real contracts and finds the problems that cost companies money.

## Key takeaways

- PowerLaw was founded in 2017 by a team from Tsinghua University's computer-science department and builds AI tools for contract review and management.

- In August 2023 it released PowerLawGLM, a hundred-billion-parameter (千亿参数) Chinese legal vertical model, jointly developed with Zhipu AI (智谱AI) on top of the ChatGLM-130B base.

- Its flagship review tool, MeCheck (智能合同审查系统), is deployed at what the company describes as nearly 100 large enterprise clients across internet, manufacturing, retail, finance, and healthcare.

- The company closed a roughly RMB 80 million (≈ US$11M / HK$86M) Pre-B round in December 2023, led by BlueRun Ventures (蓝驰创投), with Zhipu AI and others returning.

- In 2025 it pushed from contract review into delivered legal work, launching an AI lawyer agent, "吾律", aimed at small and mid-sized businesses.

## Why a legal model, not a general one

General large models are fluent but unreliable on law. The well-known failure mode is hallucination: a US lawyer was caught submitting court filings that cited six fabricated cases generated by a chatbot. PowerLaw's bet is that law rewards precision over eloquence, so it trained a model on the structure of legal text rather than treating law as just another chat topic.

The company says its corpus draws on more than 140 million adjudicated-document records (裁判文书), over 3 million regulations, and about 1 million contract files. Those are company-reported figures, not independently audited, but they describe the scale of data a legal specialist model needs to be taken seriously by corporate legal teams.

The technical spine is PowerLawGLM, built in three layers: a base model incrementally trained on legal text (LawGLM-130B), a dialogue layer fine-tuned on roughly one million legal question-answer pairs, and an application layer with engineering guards that keep citations to live statutes and reduce fabricated case law.

## What the product actually does

MeCheck does not write a legal opinion. It reads a contract, extracts key clauses — payment terms, liability, breach, dispute resolution — and marks the ones that are missing, one-sided, or risky, then proposes edits. The company says the system covers close to 1,000 review points and has been annotated on more than 10 million contract data items, and that it lifts review efficiency by over 50% on average for clients.

The companion product, MeFlow (智能合同管理), wraps review into the full contract lifecycle: drafting, approval, signing, performance tracking, and renewal. For small businesses that cannot afford in-house counsel, the company offers "无忧吾律", a low-cost legal-service brand, and in September 2025 introduced "吾律", an AI lawyer agent that can draft demand letters, legal opinions, and litigation documents and push them toward actual delivery.

According to Li An, chief scientist at BrainNet (脑机网), the meaningful shift in legal AI is from answering questions to delivering finished, file-ready documents — but the liability for the content still has to sit with a human lawyer, not the model.

Li An's comments reflect his own views and do not represent NeuroAI's editorial position or an endorsement by BrainNet (脑机网).

## The business case

Enterprise contract management is a quiet, high-volume pain point. Manual review is slow and error-prone; compliance and risk teams are thin. PowerLaw's pitch is that repetitive first-pass review can move to software, freeing lawyers for judgment calls. The company reports growth above 200% in 2023 versus the prior year and a client list spanning internet, manufacturing, retail, finance, and healthcare.

The December 2023 Pre-B round, led by BlueRun Ventures with returning investors including Zhipu AI, Sequoia China (红杉中国), Source Code Capital (源码资本), and Linear Capital (线性资本), plus e-signature vendor e签宝, gave the company room to push the model into agent-style delivery. That funding puts it among the better-capitalized specialist legal-AI players in China.

## Honest limitations

Most of the performance numbers here — 140 million documents, 1,000 review points, 50%+ efficiency gain, nearly 100 enterprise clients — come from the company itself and were not independently verified for this article. Legal AI carries real downside risk: a missed clause or a fabricated citation can create liability, so these tools are decision-support, not a substitute for qualified lawyers, and regulatory treatment of documents produced by such tools varies. We did not assess accuracy on specific contract types, data-privacy compliance for client contracts, or how the model performs against adversarial or non-standard documents. The funding figure is the company's announced round; we did not confirm post-money valuation.

## What readers can do now

- If you run contracts at a company, pilot an AI review tool on a narrow, high-volume document type (NDAs, procurement terms) before trusting it on bespoke agreements.

- If you are a small-business owner, look at fixed-fee AI legal-service products for routine letters and template drafting, but keep a human review on anything with real money at stake.

- If you evaluate legal AI vendors, ask for audit evidence on citation accuracy and hallucination rates, not just demo fluency.

---

Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-app-powerlaw-legal
Free to quote with attribution and a link to the original.
