Checkpoint-Based Governance Assessment of 2,000 AI/ML Repositories

Governance maturity assessments of 2,000 AI/ML repositories using a checkpoint-based framework. The dataset comprises two cohorts: 1,100 general AI/ML repositories and 900 high-risk repositories classified under EU AI Act criteria (healthcare, finance, LLM/GenAI, fairness/ethics). Key findings:- 94% of repositories score Critical or Low governance (below 400/1000)- 99.7% lack model cards- 99.7% lack data cards- 98.6% lack fairness testing- Healthcare AI scores lowest (171/1000) despite highest regulatory requirements Dataset includes:- Full assessment results for 2,000 repositories (CSV)- Gap analysis reports (merged and separate)- Checkpoint schema (2161 checkpoints across 7 governance domains)- README with methodology本数据集针对2000个人工智能/机器学习(AI/ML)仓库开展了基于检查点框架的治理成熟度评估。数据集包含两个分组：1100个通用AI/ML仓库，以及900个符合《欧盟人工智能法案》(EU AI Act)分类标准的高风险仓库，覆盖医疗健康、金融、大语言模型/生成式人工智能(LLM/GenAI)、公平性与伦理等领域。 核心研究结论如下： - 94%的仓库治理评分处于关键或低等级（低于400/1000分） - 99.7%的仓库未配备模型卡片(model cards) - 99.7%的仓库未配备数据卡片(data cards) - 98.6%的仓库未开展公平性测试 - 尽管医疗健康领域人工智能的监管要求最高，但其治理评分最低，仅为171/1000分 本数据集包含以下内容： - 2000个仓库的完整评估结果（CSV格式） - 差距分析报告（含合并版与独立拆分版） - 检查点框架（覆盖7个治理域，共计2161个检查点） - 包含研究方法说明的README文档