Notes, case studies and research on games, AI, data and intellectual property.
Anti-cheat decisions may rely on technology, but a permanent ban must also withstand scrutiny of the evidence. A Beijing Internet Court case illustrates what matters: the terms, server logs, appeals and account restoration.
Adapting Xiaomi-OCR-0 for local inference with NInfer, with performance tests on two GPUs, real-world OCR examples, and deployment instructions.
An employer-side lawyer reflects on a labor dispute handled from arbitration through appeal: lawful dismissal for misconduct, paid leave arrangements, and an award of additional unused annual leave pay.
An interpretation of the Supreme People's Court's 'Opinions on the Lawful Adjudication of AI-Related Dispute Cases', analyzing its impact from the perspectives of developer liability, personal data training, burden of proof for training data, open-source software liability, review of generation processes, and AI-assisted litigation documents.
Using uncensored and abliterated models and red-team datasets on Hugging Face as examples, this article explains how uncensored local models work, how they are used in red teaming, and the legal boundaries around creating, deploying, and using them.
An analysis of the possible legal risks under Chinese law—including portrait rights, human dignity, public-security penalties, and criminal liability—when someone generates sexual images or videos from a coworker’s photo without consent, even without sharing them.
A close look at the MiniMax H3 open-weight license, including copyright, revenue sharing, the US$20 million authorization threshold, territorial restrictions, prohibited uses, and how it compares with Kimi K3, GLM 5.2, DeepSeek V4, and the soon-to-be-released Qwen 3.8.
A Beijing algorithm engineer deleted roughly 89 TB of AI training data and several text-to-3D models to use company compute for freelance work. He was sentenced to five years and ten months for sabotaging a computer information system. This article examines the conviction, criminal threshold, and calculation of losses.
WordOllama, a Word/WPS add-in that I've been working on on-and-off for two years, is finally officially open-sourced today!
With big data on income tax and social insurance now in full effect, the traditional practice of 'paying social insurance at the minimum base rate' is no longer viable. This article analyzes social insurance contribution base calculation rules, whether employers can deduct back payments from salaries, and whether employees can file labor arbitration claims.
Explores how online impersonation and sarcastic content, when taken literally by AI, pollute training corpora and analyzes the legal infringement risks and negative impact on the information ecosystem.
Analyzes three types of AI large model API reverse proxies (rule abuse, payment fraud, and protection breakthrough), explores criminal regulation paths such as the crime of destroying computer information systems, and advocates for upholding the principle of criminal restraint while adopting a cross-cutting criminal-civil rights protection strategy.
Compares attorney-client privilege with AI platforms' legal obligations, analyzing whether AI platforms have confidentiality duties and reporting obligations after users confess to crimes, noting that domestic AI platforms have no confidentiality privilege whatsoever.
Analyzes whether companies can legally terminate employees without compensation for serious misconduct or causing substantial damage when they insert sensitive topics into marketing materials leading to PR crises, and the responsibility of approving managers.
Critique of AI-driven service-sector overcapacity disconnected from consumer reality, analyzing how AI layoffs eliminate consumers, one-person companies become unemployment buffers, and the digital world's new-era subsistence farming dilemma.
Introducing the WordOllama 2.0 update, featuring the new Agent panel, SKILL import, and silent review functionality, enabling Word/WPS with AI-powered autonomous contract editing, document review, and local model support.
Introduces a free Chinese law verification SKILL that lets AI agents check the accuracy of legal provisions during reasoning, preventing AI hallucinations, with installation instructions and a SKILLHub link.
Argues that the subject of software copyright protection should shift from micro-level code to the software's overall architecture, critiques the Copyright Center's zero-tolerance policy against AI-assisted code, and proposes a path of classified registration and mandatory disclosure.
Points out that after authorities standardized 'Token' as 'Ciyuan' in the AI field, translation errors arise in different contexts such as security tokens, blockchain tokens, and game tokens, providing terminology guidance for legal professionals.
Critiques the Copyright Center's requirement for software copyright applicants to pledge no AI usage as disconnected from industry reality, analyzing its impact on one-person companies, malicious reporting, and the pressure it puts on developers to collectively lie.
Analyzes the legal issues when an endorser's other controversial collaborations lead to player boycotts, noting that traditional contracts can hardly cover such risks, litigation faces evidence-proving difficulties, and commercial negotiation for settlement is the better solution.
Explores whether companies providing AI tools without token quotas constitute a failure to provide working conditions under Article 38 of the Labor Contract Law, analyzes whether employees can be fired for refusing to use AI at their own expense, and the data leak risks of self-funded API purchases.
A technical analysis of the Huawei account SDK bug caused by a NULL default value in the database that mistakenly flagged all channel server players as minors, along with a discussion of player compensation possibilities and the risks of platform SDK binding.
Using the OpenClaw craze as a starting point, this article explores how compliance requirements are seen as obstacles to rapid business growth in the AI wave, reflecting on the conflict between compliance and business in the traffic-first era.
Analyzes the legal liability of a Delta Force streamer who live-streamed the destruction of a viewer's rare in-game items, covering infringement determination, virtual property valuation methods, and the criminal threshold for intentional destruction of property.
Using the HoYoverse v. Nike Girl case as an example, this article analyzes the criminal risks of game leaking in China and seven strict causes of action in US civil litigation, revealing cross-border rights enforcement strategies.
An analysis of the eight-department joint measures for identifying minors-protecting online platforms, explaining the quantitative thresholds (minor MAU over 100K triggers special obligations) and enterprise compliance strategies including proactive reporting, self-audit, and opt-out mechanisms.
An analysis of the severe payment bug caused by player token confusion on the launch day of Arknights: Endfield's global open beta, exploring the technical causes and potential penalties under EU, US, and Japanese legal frameworks.
A critique of Generative Engine Optimization (GEO), arguing that it essentially pollutes AI-generated information by manufacturing and dumping garbage content, warning of its harm to the information ecosystem and users.
Using the incident of a software company firing employees for questioning the suit requirement at the annual meeting as a starting point, this article analyzes the reasonableness of mandatory dress codes, standards for determining disciplinary violations, and compensation calculations for unlawful termination.
Based on Tencent's open-source HY-MT 1.5 edge translation model, developed the Android local translation app PrivaTrans, covering installation, model loading, terminology assistance, and Word import/export features.
Analyzes the technical principles and impact scope of MongoDB high-risk vulnerability CVE-2025-14847 (Mongobleed), providing enterprises with remediation and emergency response plans, and explains the legal consequences of not patching under China's Cybersecurity Law, Data Security Law, and Personal Information Protection Law.
Interprets the Guangzhou Municipal Government's 'Eighteen Measures to Support the Development of the Gaming and Esports Industry,' covering pre-research grants, long-term operational rewards, technology transformation, and full-chain esports support.
Sharing the complete process of fine-tuning a legal-specific embedding model: based on Google's EmbeddingGemma-300M, trained on a legal provision-to-colloquial question dataset, outperforming Google and Alibaba's models in legal provision retrieval, now open-sourced.
Discovers W97M macro virus in official documents from the National Laws and Regulations Database, analyzes virus hazards and prevention measures, and outlines civil, administrative, and criminal liability for inadvertently publishing infected files.
Analyzes the Supreme People's Court case establishing that a game operator's defense of 'downstream didn't pay' is invalid, outlines the correct collection process for developers, and explains the risk of parent-subsidiary joint liability.
An analysis of Guangzhou's notice on coordinating annual leave during the 15th National Games, examining the legal ownership of annual leave scheduling rights, citing court cases showing companies can arrange leave after considering employee preferences, with practical recommendations.
Using the rumor that Xiaomi lost the prepayment case on appeal as an example, this article analyzes how AI web search, lacking independent judgment, can become a rumor amplifier, highlighting the importance of information discernment in the AI era.
Details the administrative, civil, and criminal liabilities under Chinese cybersecurity and personal information protection laws that game developers face if they fail to patch the high-risk CVE-2025-59489 privilege escalation vulnerability in Unity, including fines up to 5% of annual turnover and potential imprisonment for responsible personnel.
Details an AI-powered legal pleading generation platform that won second prize at the Guangdong Bar Association's AI Innovation Competition, featuring element-based form filling, multimodal OCR, smart analysis, and private model support for confidential legal document drafting.
Analyzes whether posting promotional comments under WeChat Moments paid ads constitutes infringement under China's Anti-Unfair Competition Law, distinguishing between harmless banter, free-riding commercial exploitation, and potential commercial defamation risks.
Explores whether Chinese labor law allows employers to terminate employees for cursing at the boss, examining statutory requirements for internal rules, court case precedents, and circumstances where such dismissal constitutes unlawful termination requiring 2N compensation.
Introduces an automated tool using PaddleOCR and FFmpeg to extract chat record screenshots from video or long images and generate properly formatted PDF evidence files for litigation, eliminating manual screenshot segmentation and layout work.
Reviews China's Qinglang campaign against AI technology abuse, analyzing the two-phase crackdown on unregistered AI services, deepfakes, and AI-generated rumors while questioning the feasibility and overbreadth of restrictions on training data and AI-assisted writing.
A plain-language analysis of the Supreme People's Court's ten-article Notice on standardizing enterprise-related case trials, covering equal protection, distinguishing economic disputes from crimes, and curbing profit-driven cross-regional law enforcement to foster a law-based business environment.
China's Supreme People's Court reply (2025) provides standardized interest rates for overdue foreign currency debts — using US dollar loan averages, EURIBOR, SONIA, and HIBOR — when contracts lack interest clauses.
China's 2025 judicial interpretation on IP crimes lowers prosecution thresholds for trademark counterfeiting (30,000 yuan illegal income), copyright infringement (500 copies), and trade secret theft, directly targeting knockoff games and private servers.
Analyzes Dify's modified Apache 2.0 license and argues it may violate the Open Source Initiative's definition of open source by restricting multi-tenant commercial use and prohibiting logo modification, following a public dispute where a pharmaceutical company received a lawyer letter.
A Chinese translation and summary of the EU's Key Principles on In-Game Virtual Currencies.
Breaks down the six statutory exceptions under Article 39 of China's Labor Contract Law that allow employers to legally terminate pregnant employees, clarifying that the three-periods protection is not absolute immunity.
Analyzes the first Chinese criminal case of using AI to produce pornographic novels for overseas publishing, warning that AI-generated pornographic games face the same liability under China's personal and territorial criminal jurisdiction.
Debunks widespread misconceptions about DeepSeek circulating in China's legal profession, including the myth that reasoning models always outperform regular models, that web search guarantees accuracy, and that chat input constitutes model training.
Under public pressure over aggressive data usage terms, Tencent Yuanbao revised its user agreement three times in a week, shifting from a permanent, transferable license to an opt-in model for model optimization data collection.
Explains model distillation and reveals that most locally-deployable DeepSeek-R1 versions (except 671B) are distilled from Qwen and Llama, not genuine R1, with benchmark tests showing degraded performance including circular reasoning and increased hallucinations.
An open-source elemental legal document generator that creates compliant complaint and defense templates with multi-party support, deployable locally via Python or accessed through a trial web service.
Chinese citizens publishing adult danmei content on overseas platforms face criminal prosecution under territorial jurisdiction; the gaming industry faces similar risks of cross-regional enforcement and outdated obscenity sentencing standards.
The 2024 amendments to China's Publishing Management Regulations raise minimum fines and introduce 'illegal income' as the penalty basis, with game companies facing stricter rules on which costs are deductible.
China's Qinglang special action targets algorithm-driven problems on internet platforms; game platforms must break information cocoons, respect user choice, ensure transparency, and avoid misleading algorithmic practices.
A detailed breakdown of how company equity, capital contributions, and business income are classified and divided in divorce under Chinese civil law, covering pre-marital and post-marital entrepreneurship scenarios.
Niantic's Large Geospatial Model (LGM) built from Pokemon GO player scan data raises serious legal concerns under Chinese surveying, data security, and counter-espionage laws.
Lawyers who input client information into online AI platforms risk violating confidentiality duties, as major AI services claim permanent usage rights over all user-submitted data for model training and third-party sharing.
Examines the legal interplay between non-compete agreements and consent waivers in China's tech industry, using Moonshot AI founder Yang Zhilin's HKIAC arbitration case to illustrate key risks and protection strategies for both founders and investors.
The 2024 Network Data Security Management Regulations impose heightened compliance obligations on game companies regarding data protection, minor consent, AI training data de-sensitization, and cross-border data transfer rules.
Explores Nintendo and The Pokémon Company's patent infringement lawsuit against Palworld developer Pocketpair, distinguishing patent from copyright claims, identifying potentially infringed game mechanics patents, and discussing implications for Chinese game companies expanding overseas.
Analyzes whether getting assaulted by a colleague over work feedback qualifies as a work-related injury under Chinese law, comparing two conflicting Supreme Court rulings and identifying key factors that influence the determination.
Examines how to distinguish between legitimate human body reference images and pornographic material on art employees' work computers, offering guidance for both employees and employers on policies, investigations, and wrongful termination risks.
Analyzes whether single-player game trainers for Black Myth: Wukong constitute copyright infringement, unfair competition, or the crime of destroying computer information systems, arguing they are generally legal and may even boost game sales.
Covers IP and copyright issues surrounding Journey to the West adaptations, trademark protection for Black Myth: Wukong, the legality of using real-world scanned cultural relics in games, and the risks of free-riding on the game's popularity.
An analysis of China's proposed Net ID and Net Certificate system for online identity verification, explaining how the system works, its impact on game companies' real-name authentication processes, and the privacy advantages and procedural concerns it raises.
Examines whether employers can legally require employees to reschedule missed work hours to weekends following the global 'Blue Friday' CrowdStrike outage, covering labor law provisions, wage deduction rules, and employee handbook best practices.
Explores whether using penguin imagery in games risks trademark or copyright infringement against Tencent, analyzing real examples from Chinese games and offering practical guidelines for creating original penguin characters that avoid legal exposure.
An overview of China's new Company Law capital contribution rules requiring shareholders to pay subscribed capital within five years, with specific guidance for gaming companies using subsidiary matrix structures on compliance planning, capital reduction, and avoiding penalties.
A legal analysis of a fake 'Black Myth: Wukong' cracked game package meme, examining potential trademark and copyright infringement against Game Science, unfair competition claims by miHoYo, and criminal liability for the creator and downloaders.
An overview of cross-provincial criminal case handling in China's gaming industry, covering summons, detention, arrest procedures, and company property risks.
An analysis of the 2024 Interim Regulations on Anti-Unfair Competition on the Internet and their impact on common game user acquisition practices, including shell apps, trademark misuse, fake reviews, and price discrimination, with penalty provisions.
A free offline tool for converting between Simplified and Traditional Chinese with region-specific terminology support for Hong Kong, Macau, and Taiwan.
Analyzes whether using AI tools like Midjourney and Gen-2 to recreate existing game scenes for advertising material constitutes copyright infringement, concluding that even without using original assets, riding on another game's recognition remains illegal.
Introduces an open-source offline speech-to-text tool built on Alibaba's Tongyi speech recognition model, designed for law firms and confidential industries that cannot risk uploading sensitive audio data to cloud services.
Debunks claims that Palworld's creature designs were AI-generated by examining the game development timeline against AI tool availability, revealing how internet hearsay fueled an 'AI detection' panic that replaced evidence-based criticism.
Three Chinese game companies were fined 153.49 million yuan for adding suffixes to game names, raising questions about whether common industry practices constitute illegal publication.
Tax implications of company annual party rewards in China, covering red packets, employee awards, and lucky draw prizes under personal income tax regulations.
Distinguishes between Bitcoin-style cryptocurrencies and game virtual currencies under Chinese tax law, explaining when in-game currency transactions trigger personal income tax obligations and why most casual trading remains practically untaxed.
Detailed breakdown of China's 2023 Online Game Management Measures draft, clarifying misconceptions about single-player games, ISBN rules, and minor protection provisions.
Offers a practical solution for legal professionals struggling with China's 95GB judgment document dataset by building a local SQLite database via a custom Python GUI tool, bypassing Excel's 2GB file size limit for CSV files.
An in-depth technical analysis of the Beijing Internet Court's first AI image copyright case, exploring the application and controversy of the 'originality' standard in AI-generated content.
Chinese translation of the international Guidelines for Secure AI System Development white paper co-published by CISA, NCSC, 18 countries, and 23 organizations including Microsoft, Google, and OpenAI, covering secure AI development across design, development, deployment, and operation phases.
An analysis of how video game mechanics like paid random draws, prize predictions, and chess-like gameplay may constitute operating a casino under Chinese criminal law.
Examines whether training and using AI voice models constitutes infringement under Chinese Civil Code voice rights protections, analyzing the difference between reproducing a specific person's voice versus generating composite voices from multiple sources.
A free online tool for browsing and copying Chinese legal texts, organized by category with quick navigation and regular updates for game and AI industry regulations.
Examines the legality of Unity's 2024 per-install fee model, analyzing why unilateral pricing changes are likely legal under subscription contracts, the criminal risks of using workarounds to avoid fees, and the unfair competition liability for inflating a rival's install counts.
The Long An Greater Bay Area AI Legal Research Center launched in Guangzhou, with Boyang Li appointed as Senior Advisor to advance AI legal research and compliance.
Detailed case analysis of criminal copyright infringement in China's gaming industry, including conviction thresholds, source code similarity standards, and liability for game operators.
Argues that AI copyright infringement review should center on the algorithm model's working principles, using Stable Diffusion's diffusion model to show that normally generated outputs reflect learned commonality rather than copying specific training works.
Try fewer keywords or clear the filters.