vstantch/x402-pii-corpus

--- license: mit language: - en tags: - pii - privacy - x402 - payments - synthetic - named-entity-recognition pretty_name: x402 PII Metadata Corpus size_categories: - 1K<n<10K task_categories: - token-classification task_ids: - named-entity-recognition --- # x402 PII Metadata Corpus Synthetic labelled corpus of 2,000 x402 payment metadata triples for PII filter evaluation. Released alongside the paper **"Hardening x402: Privacy-Preserving Agentic Payments via Pre-Execution Metadata Filtering"**. - **Paper:** [arXiv:2604.11430](https://arxiv.org/abs/2604.11430) [cs.CR] - **Canonical archive:** [IEEE DataPort doi:10.21227/kpsz-nq73](https://doi.org/10.21227/kpsz-nq73) - **Code:** [presidio-v/presidio-hardened-x402](https://github.com/presidio-v/presidio-hardened-x402) ## Dataset description Each record represents one x402 payment metadata triple (`resource_url`, `description`, `reason`) drawn from seven API categories. 36% of samples contain at least one synthetic PII entity injected into one of the three fields, with ground-truth span labels. ### Fields | Field | Type | Description | |---|---|---| | `id` | string | Sample identifier (`syn-NNNNN`) | | `category` | string | API category (`media`, `medical`, `financial`, `data_access`, `generic`, `compute`, `ai_inference`) | | `resource_url` | string | Synthetic x402 resource URL | | `description` | string | Payment description field | | `reason` | string | Payment reason field | | `pii_positive` | bool | `true` if any PII entity is present | | `labels` | list | Ground-truth entity annotations (see below) | ### Label schema ```json { "entity_type": "EMAIL_ADDRESS", "field": "description", "start": 24, "end": 45, "value": "<synthetic value>" } ``` Entity types: `PERSON`, `EMAIL_ADDRESS`, `US_SSN`, `IBAN_CODE`, `CREDIT_CARD`, `PHONE_NUMBER`. ### Corpus statistics | Split | Samples | |---|---| | train (all) | 2,000 | | Entity type | Count | Share | |---|---|---| | PERSON | 321 | 36.7% | | EMAIL_ADDRESS | 313 | 35.8% | | IBAN_CODE | 96 | 11.0% | | US_SSN | 85 | 9.7% | | PHONE_NUMBER | 32 | 3.7% | | CREDIT_CARD | 28 | 3.2% | ## Usage ```python from datasets import load_dataset ds = load_dataset("vstantch/x402-pii-corpus") print(ds["train"][0]) ``` ## Citation ```bibtex @misc{stantchev2026hardeningx402, title = {Hardening x402: Privacy-Preserving Agentic Payments via Pre-Execution Metadata Filtering}, author = {Stantchev, Vladimir}, year = {2026}, eprint = {2604.11430}, archivePrefix = {arXiv}, primaryClass = {cs.CR}, } ``` Corpus data also archived at IEEE DataPort: ```bibtex @misc{stantchev2026dataset, author = {Stantchev, Vladimir}, title = {Hardening x402: PII Filter Corpus, Sweep Results, and Live Ecosystem Data}, year = {2026}, publisher = {IEEE DataPort}, doi = {10.21227/kpsz-nq73}, howpublished = {IEEE DataPort, \url{https://doi.org/10.21227/kpsz-nq73}}, } ```--- license: MIT许可证 language: - 英语 tags: - 个人可识别信息（Personally Identifiable Information，PII） - 隐私 - x402 - 支付 - 合成数据 - 命名实体识别（Named Entity Recognition，NER） pretty_name: x402 PII元数据语料库 size_categories: - 1000 < 样本数 < 10000 task_categories: - 令牌分类（Token Classification） task_ids: - 命名实体识别（Named Entity Recognition，NER） --- # x402 PII元数据语料库 本数据集为包含2000条x402支付元数据三元组的合成标注语料库，用于个人可识别信息（PII）过滤器评估。本数据集与论文**《Hardening x402：基于执行前元数据过滤的隐私保护智能体支付系统》**同步发布。 - **论文**：[arXiv:2604.11430](https://arxiv.org/abs/2604.11430) [cs.CR] - **标准存档**：[IEEE DataPort 编号：10.21227/kpsz-nq73](https://doi.org/10.21227/kpsz-nq73) - **代码**：[presidio-v/presidio-hardened-x402](https://github.com/presidio-v/presidio-hardened-x402) ## 数据集说明 每条记录对应一条来自7类API的x402支付元数据三元组（`resource_url`、`description`、`reason`）。36%的样本中至少有一个合成的PII实体被注入到三个字段之一，并附带真实片段标注。 ### 字段说明 | 字段名 | 类型 | 描述 | |---|---|---| | `id` | 字符串 | 样本标识符，格式为`syn-NNNNN` | | `category` | 字符串 | API类别，可选值包括`media`（媒体）、`medical`（医疗）、`financial`（金融）、`data_access`（数据访问）、`generic`（通用）、`compute`（计算）、`ai_inference`（AI推理） | | `resource_url` | 字符串 | 合成x402资源URL | | `description` | 字符串 | 支付描述字段 | | `reason` | 字符串 | 支付原因字段 | | `pii_positive` | 布尔值 | 若存在PII实体则为`true` | | `labels` | 列表 | 真实实体标注（详见下文） | ### 标注格式 json { "entity_type": "EMAIL_ADDRESS", "field": "description", "start": 24, "end": 45, "value": "<合成值>" } 支持的实体类型包括：`PERSON`（人名）、`EMAIL_ADDRESS`（电子邮箱地址）、`US_SSN`（美国社会安全号码）、`IBAN_CODE`（国际银行账户号码）、`CREDIT_CARD`（信用卡号）、`PHONE_NUMBER`（电话号码）。 ### 语料库统计信息 | 划分 | 样本数 | |---|---| | 训练集（全量） | 2,000 | | 实体类型 | 数量 | 占比 | |---|---|---| | PERSON（人名） | 321 | 36.7% | | EMAIL_ADDRESS（电子邮箱地址） | 313 | 35.8% | | IBAN_CODE（国际银行账户号码） | 96 | 11.0% | | US_SSN（美国社会安全号码） | 85 | 9.7% | | PHONE_NUMBER（电话号码） | 32 | 3.7% | | CREDIT_CARD（信用卡号） | 28 | 3.2% | ## 使用方法 python from datasets import load_dataset ds = load_dataset("vstantch/x402-pii-corpus") print(ds["train"][0]) ## 引用格式 bibtex @misc{stantchev2026hardeningx402, title = {Hardening x402: Privacy-Preserving Agentic Payments via Pre-Execution Metadata Filtering}, author = {Stantchev, Vladimir}, year = {2026}, eprint = {2604.11430}, archivePrefix = {arXiv}, primaryClass = {cs.CR}, } 本语料库数据同时存档于IEEE DataPort： bibtex @misc{stantchev2026dataset, author = {Stantchev, Vladimir}, title = {Hardening x402: PII Filter Corpus, Sweep Results, and Live Ecosystem Data}, year = {2026}, publisher = {IEEE DataPort}, doi = {10.21227/kpsz-nq73}, howpublished = {IEEE DataPort, url{https://doi.org/10.21227/kpsz-nq73}}, }