Manifold Grace
GLOBAL AI GOVERNANCE · INFORMATION ARCHITECTURE · RESEARCH FRAMEWORK

AI Identification, Discovery & Compliance Mapping

How can governments and organisations regulate, monitor or audit AI if they cannot first establish which AI systems and uses exist? A system-of-systems methodology for locating, defining, recording and governing AI across institutions and industries.

← Primary identifiability / compliance principles · Interdisciplinary Atlas · Government & Society

The coverage problem: what is not on the map?

Primary consideration 1 — identification: Determine whether an AI system or use is present and establish its operational identity, purposes, deployment locations, relationships, accountable parties and affected groups. Primary consideration 2 — compliance: identify the applicable duties, then assess evidence of conformity, monitoring, oversight, incident response and redress.

Foundational distinction: an accurate asset inventory is the prerequisite for comprehensive organisational AI governance, but a complete inventory is difficult to prove: unknown applications, vendor-controlled components, short-lived deployments and undocumented use can remain outside the observed population. A public disclosure register is not the same as a complete internal inventory, and neither establishes that every use is compliant.

NIST AI RMF explicitly calls for AI inventories (GOVERN 1.6). The UK ATRS covers specified public-sector algorithmic uses, with scope rules and exemptions. In a written parliamentary answer on 15 July 2026, government stated that DSIT did not maintain a public or internal inventory of automated decision-making systems while explaining ATRS publication. This is evidence of a governance distinction, not a claim that every public-sector organisation lacks records.

Ontology: what exactly is an AI instance?

Use precise identity layers rather than treating AI as one undifferentiated object. A model may serve multiple products, a product may contain multiple models, and the same deployment may change versions or operators over time.

EntityDefinition for the proposed registerIdentity / evidence examples
CapabilityType of prediction, generation, ranking, optimisation or agent operationCapability taxonomy, purpose classification
ModelParticular learned artefact or model familyProvider, model ID, version, integrity digest where appropriate
AI systemApplication or service that uses one or more models and associated componentsSystem ID, purpose, architecture, owner
Deployment / runtime instanceOperating implementation in a specified environment and periodDeployment ID, effective dates, region, configuration
Use case / workflowOrganisational activity in which the system is usedProcess map, function, decision influence and affected people
Supplier / operatorEntities developing, providing, deploying or managing the systemContracts, responsibility records, service providers
Data / output / decisionInformation processed, produced or used in consequential actionData flows, decision records, output provenance
Requirement / controlApplicable legal, policy or safety obligation and its implementationJurisdiction, law, standard, control and evidence
Assessment / eventValidation, audit, incident, update, reassignment or retirementTimestamped evidence, reviewer, corrective action

The register must express many-to-many relationships: system ↔ models ↔ deployments ↔ uses ↔ suppliers ↔ people affected ↔ controls ↔ evidence. Version and time are required to avoid confusion between obsolete and current systems.

32 ways AI can go unidentified or insufficiently mapped

Each scenario is a potential discovery problem; it is not evidence of wrongdoing or undisclosed AI in any particular named organisation. Expand a card for an appropriate control.

32 scenarios

1. Embedded inference

Why it can be missed: AI is a small feature inside a larger product, with no separate user-facing label.

Identification and remediation

Inspect product architecture, supplier declarations, model dependencies and release notes.

2. Renamed conventional products

Why it can be missed: A familiar software function silently gains predictive or generative capability.

Identification and remediation

Version-by-version inventory review and vendor change notifications.

3. Feature flags and A/B experiments

Why it can be missed: AI runs only for a subset of users, sessions, territories or trial groups.

Identification and remediation

Register experiments, deployment toggles, owner and exposure criteria.

4. Remote inference APIs

Why it can be missed: A service sends data to a third-party model without visible AI branding.

Identification and remediation

Map API calls, subprocessors, contracts and service dependencies.

5. Embedded SDKs and libraries

Why it can be missed: An AI library arrives as a bundled component within another application.

Identification and remediation

Software component inventory, package audit and code review.

6. Supplier model substitutions

Why it can be missed: The supplier changes a hosted model or its underlying provider without local release.

Identification and remediation

Contractual change-control, attestations and version verification.

7. Multi-tier subcontracting

Why it can be missed: A vendor depends on another platform, model provider, retrieval service or moderation model.

Identification and remediation

Tiered supplier mapping and responsibility matrices.

8. Acquired or legacy systems

Why it can be missed: A historic application includes ML that current staff did not build or document.

Identification and remediation

Legacy asset discovery, historical procurement, architecture review.

9. Departmental shadow AI

Why it can be missed: Staff adopt unsanctioned AI applications outside approved procurement.

Identification and remediation

Safe approved-tool route, training, proportionate asset discovery and regular declarations.

10. Personal accounts used for work

Why it can be missed: Staff access public AI through individual subscriptions rather than enterprise accounts.

Identification and remediation

Policy, training, procurement checks and privacy-respecting risk review.

11. Spreadsheets and low-code workflows

Why it can be missed: AI functions hide in macros, connectors, automation platforms or extensions.

Identification and remediation

Workflow inventories, connector reviews and app owner attestations.

12. Internal ranking and triage

Why it can be missed: People see a human interface while AI prioritises cases or applications behind it.

Identification and remediation

Trace end-to-end workflows, decisions and material model contributions.

13. Decision-support presented as human judgement

Why it can be missed: An operator follows an AI score without acknowledging the model's role.

Identification and remediation

Decision mapping and evidence of meaningful human review.

14. Retrieval, moderation or safety classifiers

Why it can be missed: Background models select, suppress, route or transform content.

Identification and remediation

Component-level documentation, impact mapping, appropriate disclosure.

15. Personalisation and recommender systems

Why it can be missed: Rankings or feeds shape exposure without the AI role being obvious.

Identification and remediation

Disclose material ranking factors, document model use and complaint routes.

16. Dynamic pricing and eligibility

Why it can be missed: A model estimates price, risk, access or eligibility in ways the recipient cannot see.

Identification and remediation

Decision impact assessments, model inventories and review rights.

17. Indirect inter-agency data flows

Why it can be missed: A person encounters the decision of agency B based on AI outputs supplied by agency A.

Identification and remediation

Cross-organisation flow records, interfaces and responsibilities.

18. Agent tool calls and delegation

Why it can be missed: A visible assistant launches hidden subagents, API calls or additional models.

Identification and remediation

Agent identities, scoped permissions, parent-child task logs.

19. Autonomous scheduled processes

Why it can be missed: AI runs in unattended batches or in response to events without a direct user interaction.

Identification and remediation

Scheduler, service-account and workflow execution inventories.

20. Ephemeral serverless workloads

Why it can be missed: AI spins up briefly and leaves little visible persistent infrastructure.

Identification and remediation

Deployment manifests, signed execution records and runtime telemetry.

21. Model cascades and ensembles

Why it can be missed: One public service actually uses several specialist models that disagree or vote.

Identification and remediation

Component graph, routing and model-level version records.

22. On-device and offline AI

Why it can be missed: Inference occurs locally, without contacting a central service.

Identification and remediation

Device/vendor documentation, signed component manifests and local audit options.

23. Industrial control and machinery

Why it can be missed: ML modules operate in robotics, monitoring, production or sensor systems.

Identification and remediation

Operational technology inventories, safety cases and asset inspections.

24. Medical or scientific equipment

Why it can be missed: AI support is integrated into diagnostic, imaging or research equipment.

Identification and remediation

Manufacturer declarations, change logs and clinical/lab governance review.

25. Physical products with AI updates

Why it can be missed: An ordinary consumer product receives AI through firmware or remote updates.

Identification and remediation

Product lifecycle register, notice of material changes, manufacturer attestations.

26. Edge devices and IoT networks

Why it can be missed: Local embedded classifiers are distributed across many devices or facilities.

Identification and remediation

Device identities, fleet inventories and supplier supply-chain maps.

27. Synthetic media in mixed workflows

Why it can be missed: Content combines AI and human edits and lacks credible provenance.

Identification and remediation

Origin records and context-sensitive provenance standards, not watermark alone.

28. Proxy and outsourced decisions

Why it can be missed: An outsourced service returns only recommendations; the ordering organisation cannot inspect internals.

Identification and remediation

Contractual audit rights, purpose, output and decision traceability.

29. Cross-border hosted infrastructure

Why it can be missed: Jurisdiction, service operator and applicable obligations differ across deployment layers.

Identification and remediation

Record providers, deployment regions, data flows, user locations and legal roles.

30. Research pilots and prototypes

Why it can be missed: Small experimental AI deployments are omitted from production-only registers.

Identification and remediation

Include pilots, tests and planned deployments with status and exposure.

31. Retired but still operating systems

Why it can be missed: An AI model is declared retired but still appears in a forgotten integration.

Identification and remediation

Decommission verification, live-traffic monitoring and historical records.

32. Unclear AI boundary

Why it can be missed: Complex statistics, expert systems and advanced automation are misclassified as AI or vice versa.

Identification and remediation

Use documented definitions, classification rationale and human technical review.

Reference architecture: discovery before compliance

1 · Define scopeAI and automation classification, jurisdictions, accountable bodies
2 · DiscoverProcurement, asset inventories, approved tool lists, vendor and architecture review
3 · ReconcileDeduplicate, connect models to deployments, compare observed and declared use
4 · Identify ownersOperator, provider, business purpose and affected groups
5 · Classify dutiesRisk, sector, role, privacy, transparency and applicable law
6 · Assess complianceEvidence, audit, testing, human review, deficiencies and remediation
7 · Monitor changeNew instances, agents, suppliers, updates, incidents and retirement
8 · Report appropriatelyPublic notices where required, restricted regulator access and safe disclosure
Discovery methods and their limitations

Declarative methods: supplier attestations, procurement, contracts, employee declarations, app inventories. These are explainable but can omit hidden subcomponents and unreported use. Technical methods: authorised code/dependency inspection, cloud/service inventories, observability, endpoint agents, runtime records and security monitoring. These depend on permissions and coverage; network observations alone cannot reliably establish AI use. Human methods: staff interviews, practitioner workshops, user complaints and qualitative process mapping can identify otherwise undocumented uses. Reconciliation: require evidence from multiple independent sources, log uncertainty and review unresolved candidates.

Safeguards against overreach

Discovery must be lawful, proportionate, privacy-preserving and authorised. Do not equate every statistical tool with AI or use speculative signatures as proof. Avoid indiscriminate employee surveillance. Publish appropriately limited summaries, not credentials, exploit-relevant secrets, confidential model internals or personal data. Some sensitive systems may be documented in restricted oversight channels rather than open public registers.

Measure completeness; do not assume it

MeasureWhat it answersImportant qualification
Observed inventory coverageRegistered confirmed systems ÷ confirmed systems observed by a defined independent audit methodThis is audit-sample coverage, not proof of total real-world coverage.
Discovery precisionConfirmed AI candidates ÷ reviewed candidate detectionsDepends on the operational AI definition and reviewer quality.
Discovery recallKnown AI instances detected ÷ independently established known AI instances in a test populationUnknown unseen instances make global recall unknowable without assumptions.
Owner completenessIdentified deployments with verified accountable ownerAssigning a name without confirmed authority is not enough.
Dependency completenessVerified supplier/model relationships relative to test sampleVendor non-disclosure and nested outsourcing limit visibility.
Regulatory classification coverageIdentified uses with recorded jurisdiction, risk and applicable dutiesClassification needs legal review and may change with law.
Update latencyDelay from system change to verified inventory updateConsider short-lived or frequently changing instances.
Public comprehensibilityProportion of affected users correctly understanding a noticeUse accessible, user-tested language; not disclosure alone.
Compliance evidence coverageIn-scope controls supported by current acceptable evidenceEvidence existence is not automatically evidence of compliance.

Fundamental research gap: no inventory can prove completeness simply by counting what it already contains. Independent discovery, denominator assumptions, sample-based audits, adversarial testing and uncertainty bounds are essential. A detection tool can miss the very AI systems being sought; multiple methods reduce, but do not eliminate, blind spots.

18 associated research arenas

1. Information architecture

Faceted navigation, findability, content modelling, cross-references and public-facing maps.

2. Ontology engineering

Explicit entity definitions and machine-readable relationships: model, system, instance, use, actor, evidence, control.

3. Knowledge graphs and semantic interoperability

Connecting multiple inventories without conflating providers, models, deployments and organisational uses.

4. Enterprise and solution architecture

Business capability mapping, process flow diagrams, component and data dependency graphs.

5. IT asset and configuration management

Asset discovery, ownership, lifecycle, updates, reconciliation, unknown and retired assets.

6. Software bill-of-materials and supply-chain assurance

Dependency composition, vendor chains, libraries, hosted services and change-of-control.

7. Cybersecurity and digital forensics

Authorised verification, integrity, telemetry, incident evidence and nonintrusive discovery.

8. Data protection and information governance

Purposes, lawful bases where relevant, retention, data flows, DPIAs, accountability.

9. AI assurance and evaluation science

Measurement of inventory completeness, detection accuracy, recall, evidence quality and false assurance.

10. Risk-based regulation and regulatory technology

Applicable duties, jurisdictions, risk levels, regulatory reporting and control testing.

11. Public policy and administration

State responsibilities, contracting, agency boundaries, public reporting and redress.

12. Human–computer interaction and accessibility

Recognisable notices, user understanding, informed choice and usable challenge routes.

13. Organisation studies and socio-technical systems

Shadow adoption, internal silos, procurement incentives, reporting culture and organisational change.

14. Digital sociology and power

Who can discover system use, who bears effects, transparency asymmetries, procedural justice.

15. Systems engineering and safety science

Hazard identification, failure modes, safety cases, validation and life-cycle traceability.

16. Science and technology studies

Categories, social construction of expertise, partial knowledge, situated practice and contested boundaries.

17. Statistics, audit sampling and measurement theory

Estimating coverage, denominator uncertainty, detection bias and unseen populations.

18. International and comparative governance

Cross-border obligations, sector regulators, interoperability, accountability boundaries.

Cross-sector and global governance map

Use the same identity architecture across national and local government, education, employment, finance, healthcare, justice, migration, social care, transport, media, infrastructure, manufacturing, research and consumer services, while recording sector-specific legal duties. The same product may be low-risk in one context and materially consequential in another.

Governance layerLead functionKey question
Product developer / supplierModel identity, dependencies and documented capabilitiesWhat precisely is being sold, and what changes?
Organisation / deployerComplete local use register, purpose, roles and technical oversightWhere is AI actually used in this organisation?
Sector regulatorRisk classification, compliance evidence and sector auditsWhich uses are within statutory remit and obligations?
Cross-government coordinationCommon vocabulary, registry interoperability and inter-agency linksWhat is overlooked between departments and contractors?
International governanceDefinition alignment, cross-border responsibility and safe reportingCan compatible registries map systems that span borders?
Public / affected individualsAppropriate awareness, explanations, challenge and redressCan a person tell when AI materially affects them?

Research questions and test programme

  1. Ontology test: Can independent experts classify identical systems consistently as AI, non-AI or indeterminate?
  2. Inventory completeness: How many previously unregistered uses does a second discovery method find?
  3. Embedded systems: Can model use be mapped from procurement, configuration and supplier records without invasive surveillance?
  4. Agent traceability: Can delegated subagents, tool invocations and model substitutions be linked to accountable actors?
  5. Notice comprehension: Do affected people understand where AI contributes to decisions?
  6. Cross-agency mapping: Can dependency graphs reveal gaps between government bodies and outsourced service providers?
  7. Regulatory mapping: Can a changing legal rule be linked to every affected recorded system and use?
  8. Uncertainty and failure: When is unknown, disputed or unverified the correct classification?
  9. Material-signature hypothesis: For AI-enabled physical products, can authenticated device or hardware evidence reliably identify a relevant component? Do not assume a universal AI-specific material signature.
  10. Operational testing: How does inventory performance change under new software versions, transient services and retired-but-live instances?

Each trial should preregister its scope, baseline inventory, independent discovery method, reviewer criteria, error rates, risk controls and correction record. No global identifiability solution is claimed to have been validated by this publication.

Original research and governance references

These sources establish existing work and regulatory precedents, not universal mandatory registration. They offer starting points for expert collaboration and critical comparison.

  1. NIST AI RMF — Govern 1.6 — Voluntary governance guidance explicitly calls for AI system inventories and their completeness review.
  2. NIST AI RMF Core — GOVERN, MAP, MEASURE and MANAGE, including inventory, monitoring and decommissioning.
  3. UK NCSC — Secure AI development — Identify assets, map suppliers, document models, prompts and data.
  4. UK NCSC — Agentic AI adoption — Secure design, deployment, operating and monitoring duties as AI systems change.
  5. UK ICO — AI governance and accountability — Responsibilities, DPIAs, risk registers, oversight and periodic audits.
  6. UK ATRS — Mandatory scope and exemptions — Official scope, exceptions and distinctions between central public reporting and other tools.
  7. UK ATRS — Detailed recording guidance — Records component models and end-to-end architecture of in-scope public-sector tools.
  8. UK Parliament — July 2026 inventory answer — Government statement on DSIT inventory status and ATRS coverage.
  9. EU AI Act — Article 49 — Specified registration duties, not a universal registry of all AI deployments.
  10. EU AI Act — Article 50 — Transparency requirements for specified AI uses with exceptions.
  11. ISO/IEC 42001 — AI management systems — Cross-sector AI management-system standard; does not make all AI publicly registrable.
  12. W3C PROV — Provenance model — Entities, activities, agents, versions and derivation, useful for attribution architecture.
  13. OWASP GenAI — Security solutions landscape — Community risk/solution mapping across the AI software lifecycle.
  14. OWASP GenAI — Incident response guide — Dynamic AI-asset inventories and operational incident context.
  15. NIST — AI measurement resources — Evaluation, verification and validation resources and implementation references.
  16. NCSC — Future AI ecosystem security — Research agenda emphasising systems interacting across models, plugins, operators and organisations.

16 direct sources · Published 5 November 2026 · Curated research agenda with proposed control architecture, not a regulatory opinion, systematic review or proof of completeness.

Preserving developmental history

The earlier two-question governance framework remains intact. This page adds overlooked deployment scenarios, information-architecture ontology, inventory methods and test criteria. Changes should be date-stamped and linked, not silently erase prior proposed hypotheses or externally authored contributions.

Related: Research Atlas · Research Method · Research Gaps · Cross-sector verification.