# RiskAtlas Global Sanctions and Export Controls Database: Construction and Capability Report URL: https://cufe-digital-finance-center.vercel.app/en/reports/riskatlas-database-capability-report-2026 Published: 2026-07-20 Updated: 2026-09-06 ## APA 中央财经大学国家财经战略研究院经济制裁和国别风险研究团队. (July 20, 2026). RiskAtlas全球制裁与出口管制数据库建设与能力报告. https://atlas.scholarforce.ai/docs A capability report from NAFES describing the governance framework, data quality system, production acceptance benchmarks and AI-native services of the RiskAtlas global sanctions and export-control database, developed with research and engineering support from CUFE AI²F. ## Provenance The RiskAtlas Global Sanctions and Export Controls Database: Construction and Capability Report was published by the National Academy of Financial and Economic Strategy (NAFES) at Central University of Finance and Economics, with research and engineering support from the CUFE Institute for AI & Digital Finance (CUFE AI²F) and the Economic Sanctions and Country Risk Research Team, an independent research group of NAFES. The report was first published on 20 July 2026 and has since been revised as the database's capabilities have developed; this record reflects the version anchored to the certified data snapshot of 31 August 2026. RiskAtlas is a data platform for research on global sanctions, export controls and cross-border restriction measures. It organises official public restriction information published across the United States, the United Kingdom, the European Union, the United Nations and other jurisdictions, supporting entity search, source verification, research analysis and machine-readable access. ## Database scale and coverage As of the snapshot underlying the report, RiskAtlas held 124,527 official source records, of which 73,963 were current and 50,564 held historical status, spanning 24 issuing jurisdictions and 38 continuously maintained sources. The report treats the source record as the database's basic unit of measurement — a normalised entry within a single official list or notice. Because a real-world subject may appear across multiple jurisdictions or lists, total record counts describe source coverage rather than a deduplicated count of globally unique restricted subjects. ## Governance framework The report organises its governance approach around source priority and context preservation, field normalisation, temporal layering, tiered evidence for candidate linkage, and industry and strategic-topic tagging. Official or government-operated channels serve as the primary basis for listing facts, with issuing authority, jurisdiction, list name, status and official links preserved on each record. Name matches alone are treated as candidate leads rather than identity determinations; only corroborated formal events support a "revoked or lapsed" status, and mere absence from a current list is not automatically read as revocation. For organisational records, primary industry is tagged under UN ISIC Rev.5 with independently maintained strategic-topic labels (including semiconductors, unmanned systems, critical minerals, aerospace and defence technology, artificial intelligence, quantum technology and cybersecurity); both types of tagging require sufficient evidence and are withheld rather than guessed when evidence is insufficient. ## Data quality framework The report describes quality management across completeness, traceability, consistency, uniqueness, timeliness and statistical interpretability, introducing separate "source availability" and "parsing fidelity" metrics to distinguish what official sources actually disclose from how accurately the platform processes that disclosed information — avoiding the misclassification of upstream structural gaps as platform processing errors. ## Production acceptance baseline (31 August 2026) The report discloses production acceptance results for RiskAtlas API 1.9: 217 of 217 frontend and contract tests passed, 34 of 34 functional checks passed, and all seven API access-control scenarios passed. A 100,000-record peak screening test reached approximately 5,857 records per second with a 1.22-second p95 request time; a one-hour sustained run of 100,000 screenings recorded a 711 ms p95, with zero request errors and no consistency mismatches between API responses and underlying database records. The report notes that these figures describe controlled acceptance testing under specific conditions and do not constitute a customer-specific service-level commitment; institutional users are advised to complete shadow screening and service-level validation with their own historical samples before production use. ## AI-native service capabilities The report describes a stable API base (`/api/v1`) documented with OpenAPI 3.1, with six independently versioned dimensions — API major version, response contract version, service build version, data release version, taxonomy version and classifier version — plus an MCP (Model Context Protocol)-compatible read-only tool endpoint that organises entity search, record retrieval, status timelines, source verification, statistics and industry analysis for use by language models and agents. A `response_mode=compact` parameter is available to return a reduced field set for language-model consumption. ## Citation and usage boundaries The report's recommended database citation reads: Economic Sanctions and Country Risk Research Team, National Academy of Financial and Economic Strategy, Central University of Finance and Economics. RiskAtlas Global Sanctions and Export Controls Database [DB/OL]. [access date]. https://atlas.scholarforce.ai/. The report states explicitly that name overlaps, candidate linkages and aggregate statistics do not constitute legal-entity determinations, transaction conclusions or legal advice, and that the legal effect of any listing follows the issuing authority's official documents and subsequent amendments. Citations of specific figures should state the statistical unit, status scope, jurisdiction or source scope, and the snapshot date. Corporate author: 中央财经大学人工智能与数字财经研究中心 Original source: https://atlas.scholarforce.ai/docs ## methodology 报告基于2026年8月31日认证数据快照及同期生产验收测试结果,说明数据库的来源治理框架、字段规范方法、候选关联证据分级、行业与战略主题分类方法,以及涵盖前端契约测试、功能验收、API访问控制、峰值与持续负载测试的验收设计。 ## limitations 报告明确来源记录数量反映来源覆盖规模,不等同于去重后的现实世界受限制主体数量;行业与战略主题分类、法人识别参照和名称重合统计均为研究与检索辅助线索,不构成法律主体认定;峰值吞吐等验收指标为特定受控测试条件下的结果,不构成面向具体客户的服务等级承诺。 ## Sources - [RiskAtlas文档中心:RiskAtlas全球制裁与出口管制数据库建设与能力报告](https://atlas.scholarforce.ai/docs) - [RiskAtlas能力报告:《RiskAtlas全球制裁与出口管制数据库建设与能力报告》发布](https://atlas.scholarforce.ai/news/riskatlas-database-report-2026) - [RiskAtlas全球制裁与出口管制数据平台](https://atlas.scholarforce.ai/)