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.
从 Hugging Face 上的 uncensored、abliterated 模型与红队数据集出发，解释破限本地模型的技术原理、红队用途，以及制作、部署和使用这类模型的法律边界。
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.
从肖像权、人格尊严、治安处罚与刑事责任等层面，分析未经同意用同事照片生成色情图片或视频，即使不传播、只自己观看，可能面临哪些法律风险。
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.
从著作权、商业分成、2000万美元授权门槛、地域限制和禁止用途等方面，拆解MiniMax H3开放权重协议，并比较Kimi K3、GLM 5.2、DeepSeek V4与即将开源的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.
北京一名算法工程师为使用公司算力接外包，删除约89TB人工智能训练数据及多个文生3D模型，最终因破坏计算机信息系统罪获刑五年十个月。本文分析案件的定罪逻辑、入刑标准及损失认定。
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.
分析AI大模型API反向代理的三种类型（滥用规则、支付欺诈、防护突破），探讨破坏计算机信息系统罪等刑法规制路径，主张坚守刑法谦抑性原则并采用刑民交叉维权策略。
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.
论证软件著作权保护客体应从微观代码转向软件整体架构，批判版权中心全面拒登AI辅助代码的零容忍政策，提出分类登记与强制披露的改革路径。
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.
指出官方将AI领域Token统一译为词元后，安全令牌、区块链代币、游戏币等不同场景下Token的错误翻译问题，为法律从业者提供术语判断指引。
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.
批评版权中心要求软著申请人承诺未使用AI的政策脱离行业现实，分析其打击一人公司、催生恶意举报及逼迫开发者集体撒谎的负面影响。
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.
基于腾讯开源的HY-MT 1.5端侧翻译模型，开发安卓本地翻译软件PrivaTrans，介绍安装方式、模型载入、术语辅助和Word导入导出等功能。
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.
分享微调法律专用Embedding模型的完整过程：基于谷歌EmbeddingGemma-300M，使用法条-口语化提问数据集训练，在法条检索场景下超越谷歌和阿里同类模型，并已开源发布。
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.
从技术角度深入分析北京互联网法院AI图片著作权第一案，探讨'独创性'标准在AI生成内容认定中的适用与争议。
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.
全文翻译并解读CISA和NCSC联合发布的《安全AI系统开发指南》，为企业AI系统开发提供全生命周期的安全建议。
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.
从AI歌手翻唱到AI语音合成，系统分析AI语音技术在著作权、表演者权和人格权方面引发的各类侵权法律问题。
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.
分析扩散模型在AI创作中的侵权风险，提出算法模型应成为AI著作权认定核心的观点，探讨法律与技术融合路径。