AI Model Governance (2026): Model Risk, Registries, Model Cards & Tools

1. What Is AI Model Governance?

10 entries

Core concepts: owning, documenting, validating and monitoring models across their lifecycle. Model governance applies to every model, from classic ML to generative AI.

01NIST AI Risk Management FrameworkGovern, map, measure and manage functions for AI risk, applied to models. 02NIST AI 600-1: Generative AI ProfileRisks specific to generative AI models and suggested actions. 03ISO/IEC 42001: AI Management SystemCertifiable management system standard for organisations that build or use AI. 04ISO/IEC 23894: AI Risk ManagementGuidance on managing risk across the AI lifecycle. 05OECD AI PrinciplesIntergovernmental principles for trustworthy AI. 06Model risk managementOverview of the risk that decisions based on a faulty model cause loss. 07Google: ML Test Score and Responsible AI PracticesGoogle's practices for fairness, interpretability, privacy and safety. 08Microsoft Responsible AI StandardMicrosoft's principles and approach to responsible AI. 09Anthropic Responsible Scaling PolicyA frontier developer's framework tying safeguards to model capability. 10AI Governance FrameworksRegulation, standards, roles and controls on this site.

2. Model Risk Management and Regulatory Guidance

9 entries

Supervisory guidance that financial and other regulated firms apply to models. SR 11-7 remains the reference for validation and effective challenge.

11Federal Reserve SR 11-7: Guidance on Model Risk ManagementUS supervisory guidance on model development, validation and governance. 12Bank of England PRA SS1/23: Model Risk Management PrinciplesPrinciples for banks, covering AI and machine learning models. 13EBA: Machine Learning for IRB ModelsEuropean Banking Authority work on machine learning in regulatory models. 14EU AI Act: Regulatory FrameworkRisk-based rules, including obligations for general-purpose AI models. 15EU AI Act: General-Purpose AI ModelsGuidelines for providers of general-purpose AI models. 16EU AI Act ExplorerIndependent searchable text of the Act. 17Reserve Bank of India: FREE-AI FrameworkReserve Bank of India work on responsible AI in finance. 18MAS FEAT PrinciplesSingapore's fairness, ethics, accountability and transparency principles for AI in finance. 19Singapore Model AI Governance FrameworkPractical guidance for deploying AI responsibly.

3. Model Documentation, Cards and Inventories

11 entries

How to describe what a model is, what data trained it and where it should not be used. Documentation is the evidence regulators and auditors ask for first.

20Model Cards for Model ReportingThe paper that proposed model cards. 21Datasheets for DatasetsDocumentation for the datasets models are trained on. 22Hugging Face Model CardsGuide to writing and structuring model cards. 23Model Openness FrameworkLinux Foundation framework for classifying how open a model release is. 24CycloneDX ML-BOMBill of materials for machine learning models and datasets. 25SPDX 3.0 AI ProfileSoftware bill of materials format with AI and dataset profiles. 26The Foundation Model Transparency IndexScores developers on transparency of their foundation models. 27Anthropic Model System CardsSafety evaluations published with Claude model releases. 28Google Model Cards ToolkitOpen-source toolkit to generate model cards. 29AI Incident DatabaseCollection of real-world AI harms and failures. 30OECD AI Incidents MonitorTracks AI incidents and hazards reported in the news.

4. Model Registries and Lifecycle Management

14 entries

Systems that version models, record lineage and control promotion to production. A registry is the system of record for model governance.

31MLflow Model RegistryVersion, stage and annotate models with lineage. 32Weights & Biases RegistryRegistry for models and datasets with lineage and access control. 33Hugging Face HubHosting, versioning and gating for models and datasets. 34Amazon SageMaker Model RegistryCatalogue and approve model versions on AWS. 35Amazon SageMaker Model CardsDocument models inside SageMaker. 36Google Vertex AI Model RegistryCentral model repository on Google Cloud. 37Azure Machine Learning RegistriesShare and govern models across workspaces. 38Databricks Unity Catalog for ModelsGovern models alongside data in Unity Catalog. 39Kubeflow Model RegistryOpen-source registry for model metadata. 40DVCVersion control for data and models. 41ClearMLMLOps platform with experiment tracking and model management. 42ML Metadata Store SolutionsMetadata stores for experiments and artifacts, on this site. 43MLOps PlatformsMLOps platforms compared, on this site. 44From MLOps to AgenticOpsHow MLOps, ModelOps, LLMOps and AgenticOps stack.

5. Validation, Fairness, Explainability and Red Teaming

15 entries

Open-source and commercial tools that test whether a model is accurate, fair, robust and safe. Independent validation and effective challenge are core governance activities.

45FairlearnAssess and improve fairness of machine learning models. 46AI Fairness 360Bias detection and mitigation toolkit. 47SHAPExplain individual predictions with Shapley values. 48LIMELocal explanations for any classifier. 49InterpretMLInterpretable models and explanation tools. 50CaptumModel interpretability for PyTorch. 51GiskardTest ML and LLM models for bias, robustness and vulnerabilities. 52Holistic AIAI governance platform with open-source measurement library. 53HELM: Holistic Evaluation of Language ModelsBroad benchmark covering accuracy, fairness, robustness and toxicity. 54garakLLM vulnerability scanner. 55PyRITMicrosoft's risk identification tool for generative AI. 56promptfooTest prompts, agents and RAG, and red-team models. 57UK AISI InspectFramework for evaluating models and agents. 58MITRE ATLASKnowledge base of adversary tactics against AI systems. 59OWASP Top 10 for LLM ApplicationsMost critical risks for LLM applications.

6. Monitoring, Drift and Observability

11 entries

Detect drift, degradation and misuse once models are in production. Governance does not stop at deployment.

60EvidentlyOpen-source monitoring for data drift and model quality. 61WhyLabs and whylogsOpen-source data logging for monitoring. 62ArizeObservability and evaluation for ML and LLM models. 63FiddlerAI observability with explainability and guardrails. 64ArthurMonitoring and evaluation for ML and generative AI. 65LangfuseOpen-source tracing and evaluation for LLM applications. 66OpenTelemetry GenAI Semantic ConventionsStandard attributes for tracing generative AI calls. 67PrometheusMetrics collection for model serving. 68Amazon SageMaker Model MonitorDetects drift and quality issues in production models. 69Vertex AI Model MonitoringMonitors skew and drift for deployed models. 70AI Agent ObservabilityMetrics, events, logs and traces for agents, on this site.

7. AI Governance Platforms

12 entries

Commercial platforms that combine inventory, risk assessment, policy and compliance workflows. Compare model inventory depth, integrations and regulatory mappings.

71IBM watsonx.governanceGovernance of predictive and generative models with factsheets and monitoring. 72Credo AIAI governance platform with policy packs and risk assessment. 73ModelOpEnterprise model governance and lifecycle automation. 74Collibra AI GovernanceCatalogues AI use cases and models alongside data. 75OneTrust AI GovernanceAI inventory and risk workflows. 76SAS Model ManagerRegistry and governance for analytical models. 77MonitaurModel governance for regulated industries. 78TrustibleAI governance and compliance platform. 79AI TRiSMTrust, risk and security management for AI, on this site. 80AI Governance and Runtime Inspection SolutionsRuntime inspection and enforcement, on this site. 81AI Agent Governance SolutionsGovernance of AI agents, on this site. 82AI Context GovernanceGoverning what AI agents retrieve and remember: injection risks, access control, privacy, provenance and memory.

Frequently Asked Questions

What is AI model governance?

AI model governance is the set of policies, roles, controls and records that govern how models are built, approved, deployed, monitored and retired. It covers ownership, documentation, validation, risk assessment, access control and monitoring, and applies to classic machine learning models as well as generative AI.

How is model governance different from AI governance?

AI governance is the broader organisational framework covering regulation, ethics, roles and use cases. Model governance is the model-level layer inside it: inventory, versioning, validation, performance and drift monitoring, and change approval for each model.

What is model risk management and SR 11-7?

Model risk management controls the risk of decisions based on faulty or misused models. SR 11-7, issued by the US Federal Reserve in 2011 with the OCC, sets expectations for model development, validation, effective challenge and governance, and banks now apply it to machine learning and AI models.

What should a model inventory and model card include?

An inventory records each model's owner, purpose, version, data sources, risk tier, approval status and dependencies. A model card describes intended use, training data, evaluation results across groups, limitations and ethical considerations, so reviewers can judge fitness for use.

How do you validate an AI model before deployment?

Independent reviewers test performance on held-out data, robustness, fairness across groups, explainability and security. For generative AI this adds red teaming, prompt-injection and jailbreak tests and evaluation of hallucination and harmful outputs. Findings go back to the owner for fixes before sign-off.

What does the EU AI Act require for models?

The Act sets risk-based obligations for AI systems and separate duties for providers of general-purpose AI models, including documentation, transparency about training content and, for models with systemic risk, evaluation and incident reporting. Check the official text and timeline for dates that apply to you.

How do you monitor models in production?

Track input and prediction drift, accuracy where labels arrive, latency, cost, safety violations and user feedback. Alert on thresholds, keep traces for investigation and define when a model is retrained, rolled back or retired.

How current is this list?

Links were checked in November 2026. Some sites, such as ISO, Arize and the Fed's older pages, block automated checks or move often and may need a manual click. This page is a resource list, not legal or risk advice, and tool descriptions reflect each project's own positioning.