Nexus StandardsDigital, Cyber and AICross-cutting capabilities
Area · registry row planned

AI, models and agents

Specifies the records an AI model, statistical model or agent must carry before its outputs enter risk products: model cards, dataset sheets, evaluation records with test-set provenance, agent mandates and aiUse flags, crosswalked to the NIST AI RMF, the EU AI Act and named management-system standards. Every artifact of the area carries a UNI (Unique Nexus Identifier), which identifies and never certifies.

Area recordDigital, Cyber and AI
  1. TierCross-cutting capabilities
  2. Registryregistry row · planned
  3. Nexus StandardsNS-12, NS-10, NS-11
  4. Open standards6
  5. Feeds3 · 0 admitted
  6. Couplings4
  7. Last reviewed2026-10-17

Context entry PROPOSED for the Nexus Standards registry.

UNI
NXS-STD-DC-000003
Status
PROPOSEDregistry row · planned · scope NXQ
Last reviewed
area context bank
Cite this page

Nexus Standards (2026). AI, models and agents: area brief, Digital, Cyber and AI (PROPOSED). https://nexustandards.com/digital-ai/ai-and-models-standards/ · UNI NXS-STD-DC-000003. Last reviewed 2026-10-17.

Download the profile

Files carry the credits and the licence line; they are generated in your browser and nothing is sent. Downloads need script.

01Scope

AI, models and agents: what the area covers, and what it leaves to others.

The cross-cutting capabilities section of Digital, Cyber and AI. Scope as the area context entry states it (PROPOSED).

In scope 4

  • Model cards, datasheets and Croissant metadata for models and datasets
  • Evaluation records: benchmark, test-set provenance, metrics, subgroup results and uncertainty
  • Agent mandates, aiUse disclosure and human oversight points
  • Crosswalk to NIST AI RMF, NIST AI 600-1, EU AI Act and paywalled AI management standards as restricted profiles

Out of scope 3

  • Legal classification of a system under any AI law (a matter for counsel and competent authorities)
  • Training or hosting foundation models
  • Ranking models against each other

Jurisdictions No jurisdiction note is recorded for this area yet; jurisdiction narrows what applies and never implies that a Nexus profile satisfies a law.

02Nexus Standards that apply

3 Nexus Standards for this area.

The Nexus Standards the area context entry lists; the registry row lists NS-12, NS-10, NS-11. Every one is proposed or specified; none is adopted by an authority.

  1. 01

    NS-12 · Software, model and AI/ML artifacts (supply chain, model cards, dataset sheets)

    Status proposed · 4 rules stated. Read NS-12 · The rules for this area

  2. 02

    NS-10 · Policies and clauses (machine-readable obligations, enforcement in commons and command rooms)

    Status proposed · 7 rules stated. Read NS-10 · The rules for this area

  3. 03

    NS-11 · Verification and monitoring (evidence, attestations, zero-trust conformance observation)

    Status proposed · 6 rules stated. Read NS-11 · The rules for this area

04Open standards

6 open standards, each with its relation.

Relations are register entries (NS-14), never conformance or equivalence statements; hover or focus a relation for its meaning. Names of standards bodies and their standards belong to their stewards and are used only to identify the documents; no text of an external standard is reproduced here.

05Systemic couplings

4 couplings to other areas and subsites.

Each coupling names the mechanism that links this area to another; select a card on the canvas to trace it. Mechanisms are stated in the context bank (PROPOSED).

Canvas · the area and its couplingsDesign, not a record
AI, models and agents: systemic couplingsThe couplings the area context bank records for AI, models and agents: 1 area in Digital, Cyber and AI, 3 areas in other subsites and 3 subsites reached. Each coupling carries a mechanism stated in the context bank (PROPOSED); a reading aid, not a dependency proof.

Fig.RUNNING marks the area you are reading; DONE an area in the same subsite, QUEUED an area elsewhere and GATE a subsite reached. Tags give reading order only, never a status. Design, not a record.

  1. Data and datasetsData and EvidenceTraining and evaluation datasets are cited by UNI@version with Croissant or datasheet metadata, so a dataset correction marks dependent evaluations as stale.
  2. Identity, authorization and zero trustTrust, Verification and AssuranceAgents act only as workloads under explicit agent mandates and can file dockets but never adopt, allocate or attest.
  3. AI governance frameworks alignmentDigital, Cyber and AIModel and agent records map to the governance obligations crosswalked in the alignment area of the Digital, Cyber and AI subsite.
  4. Scenarios, models and foresightSystemic RisksSimulation and foresight models reuse the model card and evaluation record, adding scenario provenance.
06Data and feeds

3 sources, with provider, licence and delivery path.

Sources the area reads or proposes. Delivery follows GRIx first; a source is shown live only after an owner decision admits it for public pages.

Every feed · pending admission

No legal register row admits any of these sources for live display on public pages yet, so nothing is read from them here. A register row that exists records the source for the governed GRIx path; admission for public pages is a separate owner decision.

Admitted
0
none yet
Register row exists
0
recorded for the GRIx path
Proposed row
3
terms to be recorded
Delivery
0 GRIx · 1 plugin · 0 browser · 2 snapshot
GRIx first, then plugin, browser-direct, snapshot
  • AI Incident Databasepending admission

    Responsible AI Collaborative · continuous

    CC BY-SA 4.0 (incident records)

    build-time snapshotproposed register row

  • Hugging Face Hub API (model metadata)pending admission

    Hugging Face · continuous

    Hub terms of service; each model's own licence governs

    Nexus Foundry pluginproposed register row

  • OECD.AI policy observatorypending admission

    OECD · periodic

    OECD terms and conditions

    build-time snapshotproposed register row

07Indicators

2 indicators the area tracks.

Indicator names and codes as their stewards publish them; values appear only from a governed GRIx source, never typed into a page.

08Restricted compatibility profiles

2 closed standards, named only.

Closed or paywalled standards appear by name and link only; any compatibility profile of one is built on simulated data and kept on protected pages until an institutional arrangement with the issuing body is recorded. Nexus Standards is not affiliated with, endorsed by or certified by any standards body named here.

09Working group

The working-group charter, proposed.

Scope

Records for models, evaluations and agents used in risk products.

No working group is chartered yet; a community may charter this one under NS-13 through a docket.

Deliverables · 3
  1. Model card and evaluation record profile with tests
  2. Agent mandate schema and refusal rules
  3. Crosswalk table to NIST AI RMF and the EU AI Act
Liaison targets · 4
  • MLCommons liaison sought
  • OECD.AI network of experts liaison sought
  • NIST AI Safety Institute consortium liaison sought
  • CEN-CENELEC JTC 21 via a national member liaison sought
10Readiness

What each pathway can read from this area.

GCRI Canada evidence pathway

GCRI Canada

GCRI Canada uses a model's outputs as evidence only with a model card, an evaluation record on a cited test set and a stated domain of validity.

Open the GCRI Canada evidence pathway

Recognition pathway

The Global Risks Forum (GRF)

GRF recognition of an AI-enabled programme rests on recorded oversight points and incident handling, not on model capability claims.

Open the Recognition pathway

Capital pathway

The Global Risks Alliance (GRA)

GRA treats undocumented model drift or out-of-domain use as model risk to be disclosed, never as an accepted input.

Open the Capital pathway

Each institution decides under its own rules what it reads as evidence; no pathway is an approval, a readiness statement or investment advice.

11Open problems

3 open problems an expert would start with.

  1. Evaluation records for foundation models whose training data cannot be enumerated.
  2. Detecting and recording silent model updates behind third-party APIs.
  3. Making agent mandates enforceable when agents chain tools across trust boundaries.
12Audiences

Who works with this area: 4 audiences.

  1. Machine learning engineers
  2. Model risk managers
  3. AI governance and policy officers
  4. Agent and workflow designers
13Hazards and sectors

What acts on this area, and who carries it.

Hazards and threats 0

No hazard or threat is listed for this area; it reads the hazards of the areas it couples with.

Sectors 0

No sector is listed for this area; it serves every sector through the areas it couples with.

14Registry depth

The registry row: scope NXQ, 2 crosswalk domains.

This area has a row in registry/subsites.yaml (status planned), so it carries five depth pages and the register crosswalk of its domains.

Crosswalk graph · ai, models and agentsFrom the registry
AI, models and agents: the register crosswalk42 nodes on four orbits (Nexus Standards and consumers inside, then domains, external standards and issuing bodies) joined by 56 register links.
Nexus StandardsDomainsExternal standardsIssuing bodiesConsumers

3 Nexus Standards · 2 domains · 24 external standards · 12 issuing bodies · 1 consumers. Relations are register entries (adopt, profile, crosswalk, cite), never compliance claims; names belong to their stewards.

Crosswalk graph42 nodes and 56 links from registry/external-standards.yaml, registry/subsites.yaml and registry/consumers.yaml. Relations (adopt, profile, crosswalk, cite) are register entries, never compliance claims.

The registry holds no geography for the area itself: areas are not places. The one geography it implies is where the bodies that issue the area's external standards sit.

Standards globe · AI, models and agents: issuing bodiesFrom the registry
Where the issuing bodies of the external standards register sitA plate carrée world map with Natural Earth coastlines, H3 resolution 2 hexagons at the public seats of 8 issuing bodies and one dot per seat. 24 register entries.
Issuing bodies' seats (8)H3 cells, resolution 2 (5); darker, more register entriesSubsites, consumers, standards: no geography in the registry
Issuing bodies and their seats (8)
BodySeatRegister entries
ISO/IEC JTC 1/SC 42Geneva, Switzerland6
European Parliament and CouncilBrussels, Belgium4
IEEE Standards AssociationPiscataway, New Jersey, United States3
NISTGaithersburg, Maryland, United States2
Agentic AI Foundation (Linux Foundation)San Francisco, California, United States1
CEN-CENELEC JTC 21Brussels, Belgium1
ISO/IEC JTC 1/SC 7Geneva, Switzerland1
UN Office of the Secretary-General's Envoy on Technology and UNDPNew York, New York, United States1

1 subsites: no geography in the registry (areas, not places) · 2 consumers: repositories, no geography · 19 Nexus Standards: documents, no geography · 4 issuing bodies without a fixed public seat, listed not drawn: Mitchell et al. (FAT* 2019); OECD; UNESCO; United Nations · Programme regions of the ecosystem coverage register: named, not drawn (no coordinates in the registry)

Standards globe8 issuing bodies with a public seat, 24 register entries, H3 cells at resolutions 1 to 4. Seats are city coordinates, not registry fields; subsites, consumers and standards have no geography, so none is drawn. Plate carrée still map; globe with script.

Area on the sitesArtifact kindsGoverning standardsPages
Artificial intelligence (GCRI Canada field gateway; GRF and GRA counterparts)data, dataset, report, map, documentationNS-1, NS-3, NS-3b, NS-4, NS-5therisk.global/fields/ai/
Exponential risk layer: AI, compute, cyber, digital twins, biosecurity, space, quantumsoftware, model, ai-ml, documentationNS-1, proposed NS-9on the sites, page not in a published manifest

From the ecosystem coverage register: what the GCRI Canada, The Global Risks Forum (GRF) and The Global Risks Alliance (GRA) sites publish on this area, as claimed there and not evaluated here.

15References

4 sourced references.

Peer-reviewed sources and published statistics on the publishers' own domains, as the context entry cites them.

  1. peer-reviewedDatasheets for datasetsCommunications of the ACM · 2021
  2. peer-reviewedClosing the AI accountability gapACM FAT* 2020 · 2020
  3. peer-reviewedModel Cards for Model ReportingACM FAT* 2019 · 2019
  4. published statisticsArtificial Intelligence Risk Management Framework (AI RMF 1.0)NIST · 2023
16Learning

Risk Academy faculties for this area.

Plain links to the faculty pages of Risk Academy; no course, credential, recognition or endorsement is implied.

17Key terms

Terms used on this page.

UNI
Unique Nexus Identifier: identifies and resolves an artifact and its record; it never certifies truth, readiness or approval.
PROPOSED
Drafted content that no authority has adopted. Every area entry here is a proposal to the Nexus Standards registry.
crosswalk
Relation: Nexus clauses are mapped to parts of the external standard by number, never by text.
profile
Relation: a Nexus profile narrows or extends the external standard for the nexus, validated with the body's own tools where they exist.
adopt
Relation: the external standard is used as published, values validated against it.
cite
Relation: the external standard is cited by identifier and URL as a reference.
compatibility profile (restricted)
Relation to a closed or paywalled standard: named and linked only; any profile is simulated and kept on protected pages.
pending admission
No legal register row admits this source for display on public pages yet, so nothing is read from it.
GRIx
The governed data layer of Nexus Foundry: external data is read through it with its licence and source release.
docket
A public change record under NS-13, decided by editors with written reasons.
attestation
An evaluator's recorded result for one clause and one subject under NS-11; never a certification.
liaison sought
A liaison the working group would ask the body for; none is held unless the body confirms it in writing.
19Learn and apply

Where AI, models and agents is taught, worked on and run.

Risk Academy disciplines that teach AI, models and agents, with their courses; the GCRI Canada, The Global Risks Forum (GRF) and The Global Risks Alliance (GRA) subsites that carry those disciplines; and the Nexus Foundry systems the standards name. Mappings are proposed for review (Risk Academy standards map, 2026-10-17); a link says a page on the same subject exists, never an endorsement or a conformance statement; framework names in course titles are named and linked, never reproduced.

Risk Academy

Disciplines and courses

20Questions

Questions and answers.

Is the AI, models and agents area adopted?
No. The area entry is PROPOSED content for the Nexus Standards registry beside its registry row. Adoption of any standard is a recorded authority decision and none is recorded today.
Are the feeds on this page live?
No. Every feed is pending admission: no legal register row admits it for public pages yet, so nothing is read from it here.
How do I change this brief?
Through a docket under NS-13: a pull request against the area context bank or the registry, decided by editors with written reasons.
Nexus Standards

Shape this brief: AI, models and agents.

Propose a change, join the working group or name a source. Area entries are PROPOSED content for the Nexus Standards registry; a UNI identifies and never certifies, and nothing on this site is a certification, an approval or a readiness statement.

Nexus Standards documents CC BY 4.0 · tooling Apache-2.0 · counsel review pending (DEC-2026-10-16-NEXUS-STANDARDS-OPEN-LICENCE). External standards are named and linked under their own licences; a UNI (Unique Nexus Identifier) identifies, it never certifies. · Generated from the Nexus Standards corpus at commit 437ae044c5cb (2026-10-17); proposed and open.

Skip to toolbar