人工智能、模型与代理:适用的标准
登记册中列出的人工智能、模型和智能体领域的 3 项标准,每项均附有其状态、 UNI (Unique Nexus Identifier)以及文本中所述的规则。
- Nexus Standards: NS-10, NS-11, NS-12.
- 开放标准:NIST AI Risk Management Framework (AI RMF 1.0);NIST AI 600-1 Generative AI Profile;Model Cards for Model Reporting;Croissant 元数据格式;条例(欧盟)2024/1689(《人工智能法》);The Open Source AI Definition。
- 数据和数据流:AI Incident Database;Hugging Face Hub API(模型元数据);OECD.AI 政策观察站。
- Couples with: Data and datasets; Identity, authorisation and zero trust; AI governance frameworks alignment; Scenarios, models and foresight.
- Nexus Standards
- 3
- NS-10, NS-11, NS-12
- 开放标准
- 6
- 列出名称,绝不复制
- 数据和数据流
- 3
- 仅列作参考
- 耦合
- 4
- 与其他领域
软件、模型和 AI/ML 制品(供应链、模型卡、数据集说明)
Status proposed · software, models and capability lifecycle; trust, authorisation and assurance · NXS-STD-ST-000012.
| 规则 | 文本 |
|---|---|
NS12-R1 | Each software release shall carry a bill of materials, a build attestation and signatures as evidence of the UNI version. |
NS12-R2 | Each model shall carry a model card and each dataset a dataset sheet; both shall cite their sources by UNI. |
NS12-R3 | Agents shall act under recorded mandates and shall be credited as agents; they shall not adopt or certify. |
NS12-R4 | AI governance obligations shall be clauses bound to contextual UNIs and evaluated, not prose assurances. |
政策和条款(机器可读的义务:定义、约束、评估和执行)
Status proposed · trust, authorisation and assurance; integrated posture and executive synthesis · NXS-STD-ST-000006.
| 规则 | 文本 |
|---|---|
NS10-R1 | Every obligation that an implementation enforces shall be a clause with a UNI; code may implement a rule, but no rule shall be enforced without a clause record that names it. |
NS10-R2 | Subjects shall be contextual UNIs; names, labels or free text shall not bind. |
NS10-R3 | The predicate shall be data in a declared language; evaluators shall record the engine and its version; two conformant evaluators shall reach the same result for the same clause and the same inputs. |
NS10-R4 | External obligations shall be cited, not copied. |
NS10-R5 | Failure effects shall be typed; refuse 是以下对象的默认值: blocking 生产环境下的条款。 |
NS10-R6 | A clause or policy whose legalState is other than APPROVED may be evaluated, but its results shall be displayed as claims under review. |
NS10-R7 | Policies and clauses shall be append-only artefacts: withdrawal and supersession keep the record. |
验证与监测(证据、证明、零信任符合性观测)
Status proposed · trust, authorisation and assurance · NXS-STD-ST-000007.
| 规则 | 文本 |
|---|---|
NS11-R1 | A conformance statement shall rest on an attestation, and an attestation shall name an evaluator, a mandate, evidence references and a digest. |
NS11-R2 | Under zero trust, subject and evaluator shall be distinct, and production evaluations shall run outside the subject’s trust boundary. |
NS11-R3 | Missing or unresolvable evidence shall yield INDETERMINATE or REFUSED, never PASS (NS0-P12). |
NS11-R4 | Attestations shall be append-only and superseded rather than changed; monitoring records shall be observations with record UNIs. |
NS11-R5 | Every displayed conformance value shall carry its evaluation instant and its evaluator; projections shall not store verdicts. |
NS11-R6 | Exports to Verifiable Credentials, in-toto or PROV-O are crosswalks; the Nexus attestation shall remain the record of truth. |
问答。
这些就是人工智能、模型和智能体的全部规则吗?
NS1-R5 这类规则标识符表示什么?
围绕人工智能、模型与智能体。
参与制定人工智能、模型和智能体标准。
邀请政府、监管机构、标准机构、行业、学术界和社区在首个正式版本(目标为 2030 年)发布前共同完善该目录:对草案提出意见、提出变更建议、加入工作组或承办实地演练。
状态与用途。Nexus Standards 是根据 NS-13 通过公开流程制定、供自愿使用而发布的提议标准;它们不构成认证、认可或监管批准,采用与否仍由全球标准委员会以及各国的国家标准委员会决定。UNI(唯一 Nexus 标识符)用于识别制品,不对其进行认证。标准机构及其标准的名称归其管理者所有,仅用于识别相关文件;此处不转载任何外部标准的文本。Nexus Standards 与此处提及的任何标准机构均无隶属关系,也未获其认可或认证。
许可:标准文本、登记册、模式和数据采用 CC BY 4.0;参考代码采用 Apache-2.0。外部标准按名称、编号和链接引用,并适用其各自许可。第三方数据、库和字体保留其各自许可:见第三方声明。