Pith. sign in

REVIEW 4 major objections 7 minor 24 references

Durable enterprise AI value depends on a shared control plane around the model—identity, knowledge, policy, memory, tools, oversight, and evidence—not on the model alone.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-12 01:55 UTC pith:X4WPG5VG

load-bearing objection Solid vendor-neutral packaging of an emerging control-plane idea for enterprise AI; useful as architecture guidance, not as a validated system. the 4 major comments →

arxiv 2607.03516 v1 pith:X4WPG5VG submitted 2026-07-03 cs.SE cs.AIcs.CY

AGL-1: The Enterprise AI Governance Layer as a Control Plane for Trusted Enterprise Intelligence

Roopam W. Sure This is my paper
classification cs.SE cs.AIcs.CY
keywords enterprise AI governanceAI control planeagentic AIretrieval-augmented generationenterprise memorypolicy enforcementprovenanceAI observability
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

As enterprises move AI from pilots into copilots, retrieval systems, agents, and live workflows, the hard problem shifts from model access to governed intelligence operations: who may retrieve what, which policy applied, what memory was used or written, which tools ran, who approved the action, and what evidence remains. This paper introduces AGL-1, a vendor-neutral reference model for an Enterprise AI Governance Layer that sits across models, retrieval, memory, orchestration, tools, policy engines, and applications. Building on earlier governed-knowledge work, it generalizes from retrieval controls to full execution-path governance and names seven domains that address recurring failure modes such as unauthorized retrieval, stale grounding, unmanaged memory, weak provenance, policy drift, fragmented observability, and uncontrolled agent actions. The claim is that trust in complex enterprises is a system property of that managed control plane, not a property of any single model. Leaders should treat governance as reusable platform capability so teams can scale AI without reinventing authorization, evidence, and runtime limits in every app.

Core claim

AGL-1 defines the Enterprise AI Governance Layer as a design-time and runtime control plane for trusted enterprise intelligence. The paper’s central claim is that durable enterprise value from AI will increasingly depend on governing intelligence at scale: trust is not a property of the model alone but of the system around it—identity, knowledge, policy, memory, tools, human oversight, and evidence working together as a managed control plane across heterogeneous AI estates.

What carries the argument

AGL-1 (Enterprise AI Governance Layer): a vendor-neutral control-plane reference model with seven domains—identity-aware retrieval, policy enforcement, provenance management, memory governance, knowledge integrity monitoring, agentic execution control, and trust observability—that mediate the full AI execution path and produce audit-ready evidence records.

Load-bearing premise

A single vendor-neutral control-plane model with seven named domains can be implemented coherently across multi-cloud, SaaS, open-source, and internal AI estates without becoming too abstract to enforce or too fragmented to yield reusable runtime controls and evidence.

What would settle it

In a multi-vendor enterprise AI estate, measure whether adopting AGL-1-style shared controls (identity-aware retrieval, policy decision records, memory versioning, agent action traces, and linked trust observability) reduces unauthorized retrieval, policy drift, unreconstructable incidents, and unmanaged agent actions relative to siloed per-application governance; if those failure rates stay the same or rise, the control-plane claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Enterprises that treat each AI app as its own governance island will scale slower and face higher exposure than those that build reusable authorization, evidence, memory lifecycle, and observability as platform services.
  • Agentic systems will require action control—tool limits, approval gates, interrupt and rollback paths—not only output review, because failures shift from wrong language to wrong operational effects.
  • Persistent AI memory without owners, retention, correction, and version records becomes a long-lived risk amplifier rather than a productivity asset.
  • Regulated industries will compete less on model choice alone and more on whether their AI operating model can produce reconstructible lineage, policy decisions, and audit bundles at runtime.
  • CTOs can start with a minimum viable layer: inventory and risk tiers, identity-aware retrieval, policy enforcement, lineage and decision records, agent tool gates, and basic trust observability on the highest-risk paths first.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If governance evidence becomes the real product of the control plane, procurement and vendor contracts will start requiring exportable lineage, policy decision, and agent-trace schemas rather than only model APIs.
  • The same seven-domain split could become a common language for mapping existing cloud guardrails, agent gateways, and risk tools so multi-vendor estates can be scored for control-plane gaps instead of feature checklists.
  • Organizations that only instrument models and apps will still miss memory write provenance and cross-system agent trajectories; the observability gap is likely to show up first as unresolvable incident postmortems.
  • A practical next test is whether maturity scoring (0–3 per domain) predicts which AI workflows can safely move from human-in-the-loop copilots to limited autonomy without a rise in policy exceptions.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 7 minor

Summary. The paper proposes AGL-1, a vendor-neutral reference model for an Enterprise AI Governance Layer treated as a design-time and runtime control plane for trusted enterprise intelligence. Building on the author’s prior GKS-5 work on governed retrieval, it generalizes governance from RAG-specific controls to full AI execution-path concerns spanning models, retrieval, memory, agents, tools, policy, and observability. It catalogs recurring governance failure modes (unauthorized retrieval, stale grounding, unmanaged memory, weak provenance, policy drift, uncontrolled agentic execution, fragmented observability), defines seven control domains with minimum evidence requirements, specifies governance artifacts (inventory, lineage, policy decisions, memory versions, agent traces, audit bundles), and supplies a phased CTO roadmap and maturity questions. The central claim is that durable enterprise AI value depends on governing the system around the model—identity, knowledge, policy, memory, tools, oversight, and evidence—as a coherent managed control plane rather than on model capability alone.

Significance. If the framing holds, the paper offers a timely and practically useful organizing model for enterprise AI platform strategy at a moment when hyperscalers are productizing agent identity, gateways, guardrails, and governance graphs, while multi-estate enterprises still lack a shared operating model. Strengths include a clear failure-mode taxonomy (Table 3), domain-level minimum evidence lists (§6), concrete governance artifacts (Table 5), and an actionable 0–180+ day implementation roadmap (Table 6) that CTOs can use immediately. The work is honest that it is a reference model, not a product or benchmark (§12), and it usefully separates model-performance failures from authorization, memory, policy, and action-control failures. Its significance is primarily architectural and operational: a reusable vocabulary and control-plane checklist for regulated and multi-cloud AI estates, contingent on whether the seven domains can actually compose into linked runtime enforcement and reconstructable evidence rather than remaining a taxonomy.

major comments (4)
  1. The load-bearing claim that the seven domains form a coherent, vendor-neutral control plane (Abstract; §1; design principle 7 in §5.1; control-plane coherence claim in §5; Conclusion §13) is asserted rather than demonstrated. §3 documents vendor-specific building blocks (Bedrock AgentCore, Entra Agent ID, Gemini Agent Gateway, watsonx Governance Graph) but does not show composition into reusable cross-estate runtime controls and end-to-end evidence. Without at least one worked composition pattern—e.g., a single high-risk workflow with explicit domain interfaces, decision points, and a reconstructed audit path—the central claim risks reducing to a useful taxonomy. Please add a concrete end-to-end composition example or interface contracts among domains, or narrow the claim to a governance taxonomy plus operating checklist.
  2. §7 Table 5 lists governance artifacts (retrieval lineage, policy decision records, memory version records, agent action traces, audit bundles) as if they form linked, reconstructable evidence, and §6’s “minimum evidence” lists are domain-local. The paper does not specify linkage keys, required cross-references, consistency constraints, or how partial traces compose into an audit bundle under multi-vendor execution. This is load-bearing for the claim that AGL-1 produces “audit-ready evidence” and “trust observability.” Please define a minimal linkage model (identifiers, correlation IDs, retention/risk-tier rules) or a reference telemetry sketch showing how artifacts join across retrieval → policy → memory → agent action → audit bundle.
  3. §11 argues AGL-1 is especially relevant for regulated enterprises and repeatedly invokes NIST AI RMF / ISO/IEC 42001 alignment (also §3, §9, §12), but provides no control-to-framework mapping. For a paper whose audience includes risk and compliance leaders, and whose roadmap treats evidence as production-grade, a lightweight mapping of the seven domains and Table 5 artifacts to NIST AI RMF functions (Govern/Map/Measure/Manage) or ISO/IEC 42001 clauses would substantially strengthen the operating-model claim. Either supply that mapping or qualify the regulatory-alignment language as aspirational future work only.
  4. §5.2 Table 4 and §6 present seven domains as jointly necessary for control-plane coherence, but the paper does not discuss domain interactions, precedence, or failure isolation (e.g., what happens when policy enforcement blocks after retrieval has already assembled context; how knowledge integrity monitoring feeds memory governance; whether agentic execution control can override identity-aware retrieval). Without interaction semantics, implementers cannot decide enforcement order or shared state. A short subsection on domain composition rules or a control-flow diagram for a representative agentic RAG workflow would make the architecture operational rather than enumerative.
minor comments (7)
  1. Figure 1 is referenced (§5) as a conceptual control-plane view but no figure content, caption detail, or component list appears in the manuscript text provided; ensure the figure is present and that domain/enforcement-point labels match Table 4.
  2. §2 Table 1 and the GKS-5 relationship are clear, but several passages restate the same GKS-5→AGL-1 generalization (Abstract, §2, §13). Tighten repetition so the extension claim appears once with a crisp scope delta.
  3. §6.1–6.7 “Minimum evidence” bullets are valuable but inconsistently structured (some are field lists, some are outcome phrases). Normalize into a uniform schema (fields / producers / consumers / retention hint) for implementability.
  4. §9 Table 7 maturity scoring (0–3) is practical; briefly state whether scores are per-domain or aggregate and whether any domain is a hard gate for “production-grade” high-risk systems, to avoid purely additive scoring that masks critical gaps.
  5. References mix foundational ML papers ([1]–[3], [8]–[9]) with product documentation ([11]–[22]). For a governance architecture paper this is appropriate, but access dates and product-page citations should be checked for stability; prefer archival or versioned docs where possible.
  6. Minor consistency: “AGL-1” vs “the AI Governance Layer” vs “Enterprise AI Governance Layer” alternate in §10–§13; pick one primary term after first definition.
  7. §4.1 scenario is effective; consider labeling it as a running example and reusing it in §6–§7 so each domain’s minimum evidence is illustrated on the same workflow.

Circularity Check

1 steps flagged

Minor self-extension from the author's GKS-5; no derivation collapses into its inputs, and the control-plane claim is independently grounded in external sources.

specific steps
  1. self citation load bearing [Abstract; §2 Table 1; Ref. [10]]
    "Building on governed knowledge-system principles introduced in GKS-5, AGL-1 generalizes the governance problem from retrieval-specific controls to full AI execution-path governance. ... GKS-5 defines how enterprise knowledge should be governed for AI retrieval. AGL-1 defines how the broader enterprise AI environment should be governed across retrieval, memory, orchestration, policy enforcement, agentic execution, observability, and enterprise accountability."

    The paper's conceptual lineage is framed as a generalization of the same author's prior GKS-5 report rather than an independently derived architecture. This is self-citation that structures §2 and the abstract, but it is not load-bearing for the central control-plane claim: that claim is also supported by external NIST AI RMF citations and multi-vendor platform signals, so the result does not reduce solely to the self-citation.

full rationale

AGL-1 is a vendor-neutral reference architecture paper, not a fitted or first-principles derivation. There are no equations, parameters, uniqueness theorems, or empirical predictions that could reduce by construction to their inputs. The only circularity-adjacent element is structural reliance on the author's prior GKS-5 report as the retrieval-governance starting point that AGL-1 generalizes; that self-citation is normal prior-work lineage and is not the sole justification of the central claim. The seven domains, failure-mode taxonomy, evidence artifacts, and control-plane thesis are also argued from NIST AI RMF materials and hyperscaler/enterprise platform documentation (AWS AgentCore, Microsoft Agent 365 / Entra Agent ID, Google Agent Gateway, IBM Governance Graph). Industry synthesis and naming of known platform patterns as AGL-1 is the stated contribution of a reference model, not a hidden renaming of a result claimed as novel derivation. Score 2 reflects one minor non-load-bearing self-citation; the paper is otherwise self-contained as an architectural proposal.

Axiom & Free-Parameter Ledger

0 free parameters · 5 axioms · 3 invented entities

AGL-1 is a conceptual control-plane thesis. Its load-bearing content rests on domain assumptions about enterprise AI risk and operating models, plus invented named constructs (AGL-1 domains/artifacts) used to organize existing platform capabilities. There are no fitted numerical parameters or formal proofs.

axioms (5)
  • domain assumption Enterprise AI value and risk are now dominated by execution-path governance (authorization, lineage, memory, policy, agent control, evidence) rather than model access alone.
    Central premise of abstract, introduction, and conclusion; treated as the reason a distinct control plane is required.
  • domain assumption Trustworthy enterprise AI is a system property emerging from identity, knowledge, policy, runtime controls, memory lifecycle, human oversight, and evidence records interacting together.
    Stated explicitly in §4.2 and conclusion; underpins the claim that model quality alone is insufficient.
  • domain assumption Governance must be embedded in the runtime execution path and produce reconstructable evidence, not only post-deployment review or human-readable policy documents.
    Design principles 1, 4, and 5 in §5.1 and policy/provenance sections.
  • ad hoc to paper A vendor-neutral reference model can remain useful across heterogeneous multi-cloud, SaaS, open-source, and internal AI estates.
    Design principle 7 and §5.2 claim control-plane coherence without prescribing a stack; necessary for AGL-1's positioning but not demonstrated.
  • domain assumption Industry platform signals (AWS, Microsoft, Google, IBM) and NIST AI RMF indicate governance is becoming a platform layer whose integration problem remains open.
    §3 tables and citations [4],[5],[11]-[22]; used as external support for the control-plane thesis.
invented entities (3)
  • AGL-1 Enterprise AI Governance Layer no independent evidence
    purpose: Name and formalize a vendor-neutral control plane spanning models, retrieval, memory, agents, tools, policy, and observability.
    Primary invented construct of the paper; organizes existing capabilities into a coherent operating model.
  • Seven AGL-1 control domains no independent evidence
    purpose: Decompose governance into identity-aware retrieval, policy enforcement, provenance, memory governance, knowledge integrity monitoring, agentic execution control, and trust observability.
    Table 4 taxonomy is paper-defined; useful packaging but not independently validated as complete or minimal.
  • AGL-1 governance artifacts and evidence records no independent evidence
    purpose: Specify durable records (inventory, retrieval lineage, policy decisions, memory versions, agent traces, audit bundles) that make governance operational.
    Table 5 introduces a paper-specific evidence model claimed necessary for production-grade high-risk workflows.

pith-pipeline@v1.1.0-grok45 · 16596 in / 3292 out tokens · 23091 ms · 2026-07-12T01:55:55.813227+00:00 · methodology

0 comments
read the original abstract

Enterprise artificial intelligence is moving from isolated experimentation toward operational dependency across copilots, retrieval-augmented generation systems, autonomous agents, and AI-enabled business workflows. As this transition accelerates, the primary enterprise challenge is no longer only model access or inference scale. It is governed intelligence operations: the ability to enforce authorization, preserve contextual lineage, control persistent memory, detect stale or conflicting knowledge, constrain agentic execution, and produce audit-ready evidence across distributed AI estates. This paper introduces AGL-1, the Enterprise AI Governance Layer, as a vendor-neutral reference model for the control plane that should operate across foundation models, retrieval systems, orchestration frameworks, enterprise memory, policy engines, observability systems, tools, APIs, and business applications. Building on governed knowledge-system principles introduced in GKS-5, AGL-1 generalizes the governance problem from retrieval-specific controls to full AI execution-path governance. It identifies recurring failure modes such as unauthorized retrieval, stale grounding, unmanaged memory, weak provenance, policy drift, fragmented observability, and uncontrolled autonomous execution. It then defines seven governance domains: identity-aware retrieval, policy enforcement, provenance management, memory governance, knowledge integrity monitoring, agentic execution control, and trust observability. The central claim is that durable enterprise value from AI will increasingly depend on the ability to govern intelligence at scale. In complex enterprises, trust is not a property of the model alone. It is a property of the system around the model: identity, knowledge, policy, memory, tools, human oversight, and evidence working together as a managed control plane.

Figures

Figures reproduced from arXiv: 2607.03516 by Roopam W. Sure.

Figure 1
Figure 1. Figure 1: AGL-1 conceptual control-plane view. The governance layer coordinates identity, policy, provenance, memory, agent action, and observability across heterogeneous enterprise AI execution environments. 5.1 Design principles 1. Governance must be embedded into the AI execution path, not applied only as a post-deployment review. 2. Authorization must be enforced at the retrieval, memory, tool, and agent-action … view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

24 extracted references · 8 linked inside Pith

  1. [1]

    Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks,

    P. Lewis, E. Perez, A. Piktus, F. Petroni, V. Karpukhin, N. Goyal, H. Küttler, M. Lewis, W. Yih, T. Rocktäschel, S. Riedel, and D. Kiela, “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks,” in Advances in Neural Information Processing Systems, 2020. arXiv:2005.11401. AGL-1: Enterprise AI Governance Layer | Page 15

    Pith/arXiv arXiv 2020

  2. [2]

    Attention Is All You Need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention Is All You Need,” in Advances in Neural Information Processing Systems, 2017. arXiv:1706.03762

    Pith/arXiv arXiv 2017

  3. [3]

    On the Opportunities and Risks of Foundation Models,

    R. Bommasani, D. A. Hudson, E. Adeli, R. Altman, S. Arora, S. von Arx, et al., “On the Opportunities and Risks of Foundation Models,” Stanford Center for Research on Foundation Models, 2021. arXiv:2108.07258

    Pith/arXiv arXiv 2021

  4. [4]

    Artificial Intelligence Risk Management Framework (AI RMF 1.0),

    National Institute of Standards and Technology, “Artificial Intelligence Risk Management Framework (AI RMF 1.0),” NIST AI 100-1, 2023. https://www.nist.gov/itl/ai-risk-management-framework

    2023

  5. [5]

    Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile,

    National Institute of Standards and Technology, “Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile,” NIST AI 600-1, 2024. https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf

    2024

  6. [6]

    Toward Trustworthy AI Development: Mechanisms for Supporting Verifiable Claims,

    M. Brundage, S. Avin, J. Wang, H. Belfield, G. Krueger, G. Hadfield, et al., “Toward Trustworthy AI Development: Mechanisms for Supporting Verifiable Claims,” arXiv:2004.07213, 2020

    Pith/arXiv arXiv 2004

  7. [7]

    Ethical and Social Risks of Harm from Language Models,

    L. Weidinger, J. Mellor, M. Rauh, C. Griffin, J. Uesato, P. S. Huang, et al., “Ethical and Social Risks of Harm from Language Models,” arXiv:2112.04359, 2021

    Pith/arXiv arXiv 2021

  8. [8]

    GPT-4 Technical Report,

    OpenAI, “GPT-4 Technical Report,” arXiv:2303.08774, 2023

    Pith/arXiv arXiv 2023

  9. [9]

    Constitutional AI: Harmlessness from AI Feedback,

    Y. Bai, A. Kadavath, S. Kundu, A. Askell, J. Kernion, A. Jones, et al., “Constitutional AI: Harmlessness from AI Feedback,” arXiv:2212.08073, 2022

    Pith/arXiv arXiv 2022

  10. [10]

    Enterprise AI Is a Platform Problem: GKS-5, A Reference Architecture for Governed Knowledge Systems,

    R. W. Sure, “Enterprise AI Is a Platform Problem: GKS-5, A Reference Architecture for Governed Knowledge Systems,” Independent technical report, Jan. 2026. https://roopamwsure.github.io/publications/gks-5/

    2026

  11. [11]

    Amazon Bedrock Guardrails,

    Amazon Web Services, “Amazon Bedrock Guardrails,” AWS product documentation, accessed June 2026. https://aws.amazon.com/bedrock/guardrails/

    2026

  12. [12]

    Amazon Bedrock AgentCore,

    Amazon Web Services, “Amazon Bedrock AgentCore,” AWS product documentation, accessed June 2026. https://aws.amazon.com/bedrock/agentcore/

    2026

  13. [13]

    Overview - Amazon Bedrock AgentCore,

    Amazon Web Services, “Overview - Amazon Bedrock AgentCore,” AWS Documentation, accessed June 2026. https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/what-is-bedrock-agentcore.html

    2026

  14. [14]

    Add observability to your Amazon Bedrock AgentCore resources,

    Amazon Web Services, “Add observability to your Amazon Bedrock AgentCore resources,” AWS Documentation, accessed June 2026. https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/observability-configure.html

    2026

  15. [15]

    Microsoft Agent 365: The control plane for AI agents,

    Microsoft, “Microsoft Agent 365: The control plane for AI agents,” Microsoft 365 Blog, November 18, 2025. https://www.microsoft.com/en-us/microsoft-365/blog/2025/11/18/microsoft-agent-365-the-control-plane-for-ai- agents/

    2025

  16. [16]

    What is Microsoft Entra Agent ID?

    Microsoft, “What is Microsoft Entra Agent ID?” Microsoft Learn, last updated May 8, 2026, accessed June 2026. https://learn.microsoft.com/en-us/entra/agent-id/what-is-microsoft-entra-agent-id

    2026

  17. [17]

    Securing and governing the rise of autonomous agents,

    Microsoft, “Securing and governing the rise of autonomous agents,” Microsoft Security Blog, August 26, 2025. https://www.microsoft.com/en-us/security/blog/2025/08/26/securing-and-governing-the-rise-of-autonomous-agents/

    2025

  18. [18]

    Introducing Gemini Enterprise Agent Platform, powering the next wave of agents,

    Google Cloud, “Introducing Gemini Enterprise Agent Platform, powering the next wave of agents,” Google Cloud Blog, April 22, 2026. https://cloud.google.com/blog/products/ai-machine-learning/introducing-gemini-enterprise- agent-platform AGL-1: Enterprise AI Governance Layer | Page 16

    2026

  19. [19]

    Agent Gateway overview,

    Google Cloud, “Agent Gateway overview,” Gemini Enterprise Agent Platform Documentation, accessed June 2026. https://docs.cloud.google.com/gemini-enterprise-agent-platform/govern/gateways/agent-gateway-overview

    2026

  20. [20]

    Agent Identity overview,

    Google Cloud, “Agent Identity overview,” Gemini Enterprise Agent Platform Documentation, accessed June 2026. https://docs.cloud.google.com/gemini-enterprise-agent-platform/govern/agent-identity-overview

    2026

  21. [21]

    watsonx.governance,

    IBM, “watsonx.governance,” IBM product page, accessed June 2026. https://www.ibm.com/products/watsonx- governance

    2026

  22. [22]

    Governing assets with watsonx.governance,

    IBM, “Governing assets with watsonx.governance,” IBM Documentation, accessed June 2026. https://www.ibm.com/docs/en/watsonx/saas?topic=governing-ai

    2026

  23. [23]

    The 2025 AI Agent Index: Documenting Technical and Safety Features of Deployed Agentic AI Systems,

    L. Staufer, K. Feng, K. Wei, L. Bailey, Y. Duan, M. Yang, A. P. Ozisik, S. Casper, and N. Kolt, “The 2025 AI Agent Index: Documenting Technical and Safety Features of Deployed Agentic AI Systems,” arXiv:2602.17753, 2026

    Pith/arXiv arXiv 2025

  24. [24]

    ISO/IEC 42001:2023 Information technology - Artificial intelligence - Management system,

    International Organization for Standardization, “ISO/IEC 42001:2023 Information technology - Artificial intelligence - Management system,” 2023. https://www.iso.org/standard/42001

    2023