--- title: "AI is now in China's courtrooms — but who is really judging?" date: 2026-10-06 category: AI in Action site: NeuroAI canonical: https://neuroai.site/a/na-app-ai-legal-assistance language: en --- # AI is now in China's courtrooms — but who is really judging? > From Hangzhou's internet court to Shanghai's '206' evidence system, China has deployed AI across judging, contracting and policing — with real efficiency gains and open fairness questions. A judge in Suzhou dictates a verdict and the transcript writes itself. A small-business owner in Shanghai uploads a contract and software flags the clause that could sink them. The same technology that promises speed also raises a harder question: when a machine helps decide, who is accountable? China has rolled out legal AI faster and more broadly than almost any other jurisdiction — not as a single product, but as a layer stitched through courts, prosecutors, law firms and corporate legal teams. ## A court that never closes The Supreme People's Court first proposed the "smart court" (智慧法院) concept in 2016. The flagship experiment came a year later: on 18 August 2017, Hangzhou opened the world's first internet court, hearing cases entirely online. Beijing and Guangzhou followed with their own internet courts in September 2018. The efficiency was immediate and measurable. According to the official China Court network, internet-court trials averaged about 25 minutes, with an average handling period of 48 days — more than twice as fast as traditional proceedings for comparable online disputes. ## From e-filing to e-verdict The pandemic forced the system's hand. In 2020, courts nationwide handled 29.96 million cases filed online, held 5.04 million online hearings, exchanged 8.19 million batches of evidence online, and ran 5.93 million remote enforcement actions, the Supreme People's Court reported. Behind those numbers sit assistive tools. The 法信 (Faxin) platform pushes similar-case references, checks documents for errors and surfaces relevant statutes for judges. In Suzhou's intermediate court, voice recognition lifted transcript completeness to near 100% and shortened trial time by an estimated 20–30%, per the court's own account. ## The '206' system and the evidence question The most debated deployment is Shanghai's "206" intelligent case-assistance system (刑事案件智能辅助办案系统), built with iFlytek (科大讯飞) and rolled out from May 2017. Its purpose was narrow but consequential: embed unified evidence standards across the police, procuratorate and court so that flawed evidence is caught early. The Supreme People's Procuratorate described its early pilot — by 31 July 2017, 65 cases entered, 20,715 evidence items logged, 3,712 guidance prompts issued and 48 evidence defects flagged. By the end of 2018, Shanghai reported "three 100%" coverage: evidence-standard guidance across common offences, case entry into the system, and frontline use by handling officers. ## Contract law, automated Outside the state system, private legal AI is growing fast. PowerLaw (幂律智能), founded in September 2017 by a Tsinghua University computer-science team, builds contract-review tools (MeCheck, MeFlow). In 2023 it partnered with Zhipu AI to train PowerLawGLM, a legal large model (大模型) aimed at clause risk, extraction and drafting. On the court-supplier side, THUNISOFT (华宇软件) has spent more than 20 years in legal technology and says it serves over 3,000 courts nationwide. Its legal-AI platform is built on a knowledge graph of more than 170,000 legal elements, with sentencing-coverage for roughly 94% of criminal charges and about 95% accuracy on cause-of-action inference, per the company. ## Where the efficiency is real The wins are not hypothetical. Online filing removes geography. Voice transcription removes drudgery. Evidence-checking flags gaps a tired human might miss. For high-volume, repetitive matters — traffic disputes, routine contracts, standard filings — AI clearly cuts cost and delay. According to Li An, Chief Scientist at BrainNet (脑机网), China's authoritative AI observatory, the real risk is not that AI decides cases but that its quiet recommendations get accepted without challenge, because questioning a machine feels like work. ## The fairness questions nobody can skip A tool that speeds judgment also risks automating the biases already in the data. A "similar-case" engine inherits whatever inconsistency exists in past rulings. An evidence-checker bakes in whatever standard was coded. And when a recommendation is opaque, defendants and lawyers lose the ability to contest the reasoning — not the outcome. Chinese courts have framed these systems explicitly as assistive: the judge decides, the machine advises. Whether that line holds in practice depends on transparency, audit trails and the right to contest an algorithmic suggestion — safeguards that are still maturing. ## Honest limitations - Efficiency figures (25-minute trials, 20–30% time savings, "three 100%") come from official court and procuratorate communications, not independent academic replication. - THUNISOFT's 170,000-element graph and 94%/95% accuracy rates are company-disclosed and have not been independently benchmarked by a third party. - The "206" system's records describe pilots and Shanghai coverage; we did not verify rollout quality in every province where it was later trialed. - PowerLaw's founding date, Tsinghua origins and 2023 PowerLawGLM partnership are company-stated; we treated them as corporate history, not performance claims. - This article covers field deployment of legal AI; it does not assess the underlying policy or statute. ## What readers can do now - **Treat AI legal output as a draft, not a verdict** — whether it's a contract review or a case reference, have a qualified human validate the reasoning before acting. - **Ask for the audit trail** — when a court or agency uses an assistive system, transparency about what was suggested and who overrode it is the line between help and hazard. - **Use the public tools first** — China's online filing, evidence exchange and internet-court channels are real, free infrastructure; start there before paying for private legal AI. --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-app-ai-legal-assistance Free to quote with attribution and a link to the original.