Skip to content

Repository files navigation

AI Security Governance Toolkit

Practical governance artifacts for organizations adopting large language models, AI assistants, and agentic workflows in regulated, education, and public-sector environments.

This toolkit is built around a simple premise: AI security is not only a model problem. It is a governance, data, identity, continuity, vendor, assurance, and operating-model problem.

Who This Is For

  • Information security and risk leaders building AI governance programs
  • IT and digital transformation teams introducing AI-assisted workflows
  • Education technology teams evaluating AI tutors, assistants, and learning tools
  • Internal audit, compliance, and assurance teams reviewing AI control evidence
  • Researchers and practitioners translating AI security guidance into practical operations

What Is Inside

Area Artifacts
AI system intake Initial AI use-case review, data classification, deployment model, ownership, and approval path
Risk management AI risk tiering, AI risk register, risk-register reporting, control ownership, inherent/residual scoring, and decision tracking
Risk acceptance Time-bound model risk acceptance records with affected users, compensating controls, expiry dates, review triggers, approval evidence, and closure conditions
Governance forums Repeatable review agendas, decision logs, risk acceptance records, required participants, and follow-up tracking
Decision contestability Review template for notice, explanation, appeal, correction, human review, and remediation controls for AI-supported decisions
Human review calibration Template for testing reviewer independence, consistency, overrides, workload, bias indicators, escalation, and evidence quality
Emergency stop drills Template for testing AI pause, containment, rollback, fallback, evidence preservation, communications, and restart controls
Evaluation contamination review Template for checking benchmark leakage, holdout exposure, answer-key retrieval, synthetic reuse, reviewer conflict, and vendor boundary evidence
Control catalog Governance, identity, data, model, agent, vendor, monitoring, and continuity controls
Evidence management Evidence register, test records, exception tracking, and audit-ready documentation
Evidence quality Evidence freshness, ownership, source, cadence, status, and owner-review queues
Evidence retention Retention windows, storage controls, deletion triggers, legal holds, and privacy minimization for AI governance evidence
Control testing Evidence packs and examples for AI control design and operating-effectiveness review
Evaluation evidence Release-readiness reporting for AI evaluation suites, dataset lineage, security cases, human review, pass rates, and stale evidence
Exception management Aging report for expired, expiring, missing-expiration, and inconsistent AI governance exceptions
Data deletion evidence Register and report for prompt, output, file, embedding, log, vector-store, and third-party deletion proof
Vendor due diligence AI procurement scoring, vendor questionnaire, data handling, security, model controls, compliance, and continuity
Third-party dependency assurance Provider/subprocessor register, assurance review, DPA status, subprocessor transparency, exit planning, and continuity evidence
Shadow AI discovery Triage template for unmanaged AI tools, data exposure, owner assignment, remediation, and exception handling
Synthetic data governance Release review template for lineage, memorization, re-identification, utility, consent, release boundaries, and post-release monitoring
Logging and retention Checklist for prompt, output, embedding, tool-call, and operational log governance
Agentic safety Tool-use review, privilege boundaries, human approval gates, and execution logging
Agent memory governance Persistent memory register and review report for consent, stale facts, deletion evidence, retrieval boundaries, and memory poisoning risks
Agentic risk mapping Agentic failure modes mapped to controls, evidence, release gates, monitoring signals, and owner roles
OWASP LLM 2025 mapping OWASP LLM application risks mapped to toolkit controls, evidence, release gates, monitoring signals, and owner queues
NIST AI RMF crosswalk Govern, Map, Measure, and Manage functions mapped to controls, evidence, review cadence, and owner roles
Access recertification User, administrator, service account, API key, break-glass, and agent tool permission review
Red-team readiness Scenario-based AI red-team test planning, evidence capture, and release decisions
Red-team taxonomy Finding classification by risk theme, severity floor, release decision, owner role, and control coverage
Incident response AI-specific incident severity classification, escalation triggers, and evidence preservation
Incident evidence Register and report for AI incident timelines, containment evidence, preserved logs, privacy review, communications, and remediation queues
Tabletop evidence Exercise evidence register and reporting for decision logs, containment, communications, fallback, privacy review, and remediation
Decommissioning Retirement checklist for model, provider, data, credential, log, vector-store, and integration shutdown
Model/provider changes Approval workflow for model, provider, endpoint, embedding, safety-filter, region, and agent-runtime changes
Model monitoring KPI register for quality, safety, retrieval integrity, drift, security, reliability, cost, privacy, and oversight
Retrieval integrity Review template for RAG source authority, temporal validity, stale facts, citation precision, and poisoned-source controls
EdTech example AI assistant governance profile for learning platforms and student-facing workflows
Playbooks Prompt injection incident response and AI service outage tabletop scenarios
Policy as code Open Policy Agent examples for AI tool execution decisions

Reference Frameworks

The toolkit is designed to be framework-aware without pretending to replace formal assurance work.

Quick Start

  1. Copy templates/ai-system-intake.md for each AI use case.
  2. Classify the use case with templates/ai-risk-tiering-decision-record.md.
  3. Record risks in templates/ai-risk-register.csv.
  4. Select controls from controls/control-catalog.yaml.
  5. Track evidence in templates/evidence-register.csv.
  6. Schedule operating-effectiveness tests in templates/ai-control-test-schedule.csv.
  7. Use templates/agentic-tool-review.md before allowing AI agents to call tools or APIs.
  8. Score external AI products with templates/ai-procurement-scoring-worksheet.md.
  9. Complete templates/vendor-ai-due-diligence.md before approving external AI services.
  10. Review logging decisions with templates/ai-logging-retention-checklist.md.
  11. Run a tabletop exercise from playbooks/ before production rollout.
  12. Track tabletop exercise evidence with templates/ai-tabletop-exercise-evidence-register.csv.
  13. Use templates/ai-system-decommissioning-checklist.md before retiring, replacing, or pausing an AI system.
  14. Use templates/model-provider-change-approval.md before changing models, providers, endpoints, embeddings, regions, or safety filters.
  15. Track production health with templates/model-monitoring-kpi-register.md.
  16. Package control test evidence with templates/ai-control-test-evidence-pack.md.
  17. Review active exceptions with scripts/exception_aging_report.py.
  18. Summarize residual risk and owner review queues with scripts/risk_register_report.py.
  19. Review deletion evidence gaps with scripts/data_deletion_evidence_report.py.
  20. Review provider, subprocessor, and critical AI dependency gaps with scripts/third_party_dependency_report.py.
  21. Review tabletop evidence gaps with scripts/tabletop_evidence_report.py.
  22. Review AI evaluation evidence readiness with scripts/evaluation_evidence_report.py.
  23. Map agentic failure modes to controls and release gates with scripts/agentic_risk_control_report.py.
  24. Review AI access recertification gaps with scripts/access_recertification_report.py.
  25. Review AI incident evidence gaps with scripts/incident_evidence_report.py.
  26. Map OWASP LLM 2025 risks to governance controls with scripts/owasp_llm_mapping_report.py.
  27. Map NIST AI RMF functions to toolkit evidence with scripts/nist_ai_rmf_crosswalk_report.py.
  28. Summarize AI red-team finding taxonomy coverage with scripts/red_team_taxonomy_report.py.
  29. Review evidence quality and freshness with scripts/evidence_quality_report.py.
  30. Review AI data provenance, transformation evidence, transfer records, and owner queues with scripts/data_lineage_report.py.
  31. Triage AI privacy review and DPIA triggers with scripts/dpia_triage_report.py.
  32. Review rollback readiness for model, prompt, provider, safety-filter, and RAG releases with scripts/rollback_readiness_report.py.
  33. Set evidence retention, deletion, storage, and legal-hold expectations with templates/ai-evidence-retention-schedule.md.
  34. Review persistent memory with templates/ai-agent-memory-governance-register.md and templates/ai-agent-memory-governance-report.md.
  35. Run governance forums with templates/ai-governance-review-agenda.md.
  36. Review RAG, knowledge-base, vector-store, and memory retrieval behavior with templates/ai-retrieval-integrity-review.md.
  37. Triage unmanaged or unclear AI use with templates/ai-shadow-ai-discovery-triage.md.
  38. Review synthetic data releases with templates/ai-synthetic-data-release-review.md.
  39. Review appealability and human oversight for high-impact decisions with templates/ai-decision-contestability-review.md.
  40. Calibrate human review quality with templates/ai-human-review-calibration.md.
  41. Test emergency stop readiness with templates/ai-emergency-stop-drill.md.
  42. Review evaluation dataset contamination risk with templates/ai-evaluation-contamination-review.md.
  43. Record time-bound residual model risk decisions with templates/ai-model-risk-acceptance-register.md.

Validate repository artifacts before opening a pull request:

python scripts/validate_repository.py

The validator checks Markdown templates, template index coverage, control catalog structure, CSV headers, and policy-as-code examples.

Artifact Catalog

Artifact Purpose
templates/ai-change-impact-assessment.md Review governance, data, model, tool, compliance, and continuity impact after AI system changes
templates/ai-risk-tiering-decision-record.md Classify AI systems into risk tiers, set required control baselines, define approval routes, and record residual-risk decisions
templates/ai-shadow-ai-discovery-triage.md Triage unmanaged AI tools discovered through telemetry, expenses, surveys, DLP alerts, support tickets, procurement, or self-disclosure
templates/ai-synthetic-data-release-review.md Review synthetic data release readiness across lineage, privacy, re-identification, utility, consent, release boundaries, and monitoring
templates/ai-system-decommissioning-checklist.md Retire AI systems safely by handling data, credentials, vector stores, logs, vendor access, integrations, evidence, and residual risk
templates/ai-data-flow-record.md Document AI data sources, processing steps, destinations, classifications, and controls
templates/ai-decision-contestability-review.md Review notice, explanation, evidence access, correction, appeal, human review, and remediation controls for AI-supported decisions
templates/ai-emergency-stop-drill.md Test stop triggers, stop authority, containment, evidence preservation, fallback, rollback, communications, and restart readiness
templates/ai-evaluation-contamination-review.md Check benchmark leakage, holdout exposure, answer-key retrieval, synthetic reuse, reviewer conflict, and vendor boundary evidence before relying on evaluation scores
templates/ai-model-risk-acceptance-register.md Record time-bound residual risk acceptance with compensating controls, affected users, expiry dates, review triggers, approval evidence, rollback conditions, and closure status
templates/ai-human-review-calibration.md Test human review independence, consistency, override quality, automation-bias indicators, reviewer workload, and escalation evidence
templates/ai-data-lineage-register.csv Track source systems, owners, classifications, transformations, legal basis, retention, transfers, and review dates for AI data assets
templates/ai-data-lineage-report.md Operating guide for reviewing data provenance, source ownership, transformation evidence, legal basis, retention, transfer, and review freshness
templates/ai-data-deletion-evidence-register.csv Track AI deletion requests, processors, scopes, due dates, completion evidence, verification, and retention exceptions
templates/ai-data-deletion-evidence-report.md Operating guide for reporting overdue deletions, missing evidence, unverified completions, and retention-exception review queues
templates/ai-dpia-triage-register.csv Track DPIA triggers, personal-data use, automated decisioning, monitoring, transfers, privacy notices, and owner decisions
templates/ai-dpia-triage-report.md Operating guide for privacy review triggers, owner queues, and DPIA launch blockers
templates/ai-rollback-readiness-register.csv Track rollback triggers, runbooks, evaluation baselines, traffic-shift plans, schema compatibility, credentials, communications, drills, and approvals
templates/ai-rollback-readiness-report.md Operating guide for release rollback readiness and owner queues
templates/ai-access-review.md Review user, administrator, service account, API key, and agent tool permissions
templates/ai-access-recertification-register.csv Track AI access assignments, privileged roles, service accounts, tokens, tool permissions, MFA, review dates, and recertification decisions
templates/ai-access-recertification-report.md Operating guide for reviewing stale access, unowned tokens, separated users, break-glass approval, and high-impact agent tool permissions
templates/ai-agent-memory-governance-register.md Record persistent memory assets, write rules, retrieval boundaries, lifecycle evidence, deletion proof, and approval decisions for AI agents and assistants
templates/ai-agent-memory-governance-report.md Review consent, deletion, stale-fact handling, access boundaries, provider transfer, memory poisoning, and owner queues before enabling persistent memory
templates/human-approval-matrix.md Define when AI-assisted actions require review, approval, or dual approval
templates/board-ai-security-assurance-checklist.md Executive assurance checklist for approving high-impact AI systems
templates/ai-control-evidence-raci.md Assign control ownership and evidence accountability across AI governance roles
templates/ai-control-test-evidence-pack.md Package AI control test objectives, population, sample, evidence inventory, exceptions, remediation, and reviewer conclusions
templates/ai-control-test-evidence-examples.csv Spreadsheet-friendly examples for operating-effectiveness evidence packs across agent tools, prompt injection, RAG boundaries, monitoring, and continuity
templates/ai-evidence-quality-report.md Operating guide for reviewing stale, missing, expired, unowned, or incomplete AI governance evidence
templates/ai-evidence-retention-schedule.md Define AI governance evidence retention periods, storage controls, deletion triggers, legal holds, privacy minimization, and owner accountability
templates/ai-evaluation-evidence-register.csv Track AI evaluation suites, dataset lineage, model/prompt/index versions, pass rates, failures, security/bias/citation coverage, human review, and release decisions
templates/ai-evaluation-evidence-report.md Operating guide for reviewing stale, incomplete, low-scoring, or release-blocking AI evaluation evidence
templates/ai-red-team-test-plan.md Plan AI red-team scenarios, evidence capture, pass/fail criteria, and release decisions
templates/ai-retrieval-integrity-review.md Review RAG source authority, temporal validity, stale fact handling, citation precision, retrieval boundaries, and poisoned-source controls
templates/ai-governance-escalation-decision-tree.md Route AI use cases, incidents, model/provider changes, agent tools, and research risks to the correct governance review path
templates/ai-governance-review-agenda.md Run structured governance forums with required participants, pre-read evidence, decision options, risk acceptance, and follow-up tracking
controls/ai-red-team-finding-taxonomy.md Classify AI red-team findings by risk theme, severity floor, evidence, release decision, owner role, and control mapping
controls/ai-red-team-finding-taxonomy.csv Spreadsheet-friendly red-team finding taxonomy for dashboards, approval gates, and remediation queues
templates/ai-risk-register-report.md Operating guide for summarizing residual AI risk by theme, owner, and review-date status
templates/ai-incident-severity-matrix.md Classify AI incidents by data exposure, tool misuse, output harm, identity impact, continuity, and escalation triggers
templates/ai-incident-evidence-register.csv Track AI incident timelines, evidence references, data exposure, tool misuse, containment evidence, preserved logs, privacy review, communications, root cause, remediation due dates, and closure status
templates/ai-incident-evidence-report.md Operating guide for reviewing missing AI incident evidence, privacy review gaps, tool-misuse log preservation, containment proof, incomplete timelines, and overdue remediation
templates/prompt-injection-test-record.md Record direct and indirect prompt injection test cases, outcomes, and remediation
templates/ai-agent-tool-inventory.csv Track agent tools, environments, permissions, data access, owners, and review dates
templates/ai-control-test-schedule.csv Schedule AI control testing, evidence sources, test owners, results, and remediation due dates
templates/ai-exception-register.csv Track AI governance exceptions, expirations, risk owners, and compensating controls
templates/ai-exception-aging-report.md Operational guide for aging AI exceptions and converting registers into risk-owner review queues
templates/ai-model-card-lite.md Capture lightweight model use, limitations, oversight, and risk notes
templates/ai-procurement-scoring-worksheet.md Score AI vendors, hosted models, embedded AI features, and agent platforms before purchase, renewal, or expansion
templates/ai-procurement-scoring-worksheet.csv Spreadsheet-friendly companion for weighted procurement scoring, evidence gaps, and remediation ownership
templates/ai-third-party-dependency-register.csv Track AI providers, subprocessors, critical services, data access, DPA status, assurance evidence, exit plans, and continuity posture
templates/ai-third-party-dependency-report.md Operating guide for reviewing third-party AI dependency gaps before release, renewal, audit, or provider change
templates/ai-tabletop-exercise-evidence-register.csv Track AI tabletop exercise evidence, decision logs, containment tests, communications, fallback, privacy review, and remediation
templates/ai-tabletop-evidence-report.md Operating guide for reviewing tabletop exercise gaps before release, audit, or governance review
templates/model-provider-exit-plan.md Plan provider exit triggers, fallback options, data export, and deletion evidence
templates/model-provider-change-approval.md Review model/provider changes for data handling, quality, safety, RAG behavior, tool use, cost, continuity, and approval routing
templates/model-monitoring-kpi-register.md Define post-approval AI monitoring KPIs, thresholds, evidence sources, owners, and escalation rules
templates/model-monitoring-kpi-register.csv Spreadsheet-friendly companion for model monitoring KPI tracking and review records
controls/agentic-risk-control-mapping.md Map agentic AI failure modes to practical controls, evidence, release gates, monitoring signals, and owner roles
controls/agentic-risk-control-mapping.csv Spreadsheet-friendly companion for agentic risk-to-control mapping
controls/owasp-llm-2025-control-mapping.md Map OWASP LLM 2025 risks to toolkit controls, evidence expectations, release gates, monitoring signals, and owner roles
controls/owasp-llm-2025-control-mapping.csv Spreadsheet-friendly companion for OWASP LLM 2025 governance mapping
controls/nist-ai-rmf-control-crosswalk.md Map NIST AI RMF Govern, Map, Measure, and Manage functions to toolkit controls, evidence, cadence, and owner roles
controls/nist-ai-rmf-control-crosswalk.csv Spreadsheet-friendly companion for NIST AI RMF evidence crosswalks
policies/ai-log-retention-policy.md Define retention expectations for prompts, outputs, embeddings, tool calls, and security events
playbooks/ai-data-leak-triage.md Triage suspected AI data exposure through prompts, outputs, logs, embeddings, or providers
playbooks/agentic-tool-misuse-response.md Contain and investigate incorrect or unauthorized agent tool actions
scripts/exception_aging_report.py Generate Markdown or JSON reports for expired, expiring, missing-expiration, and inconsistent AI exceptions
scripts/risk_register_report.py Generate Markdown or JSON summaries of residual AI risk, owner queues, review dates, and theme-level exposure
scripts/data_deletion_evidence_report.py Generate Markdown or JSON reports for AI data deletion evidence gaps and processor review queues
scripts/third_party_dependency_report.py Generate Markdown or JSON reports for provider, subprocessor, assurance, exit-plan, and continuity gaps
scripts/tabletop_evidence_report.py Generate Markdown or JSON reports for tabletop exercise evidence gaps, owner queues, and release-blocking readiness issues
scripts/evaluation_evidence_report.py Generate Markdown or JSON reports for AI evaluation evidence readiness, stale evidence, missing coverage, release-blocking failures, and owner queues
scripts/agentic_risk_control_report.py Generate Markdown or JSON summaries of agentic risk mappings, owner queues, control coverage, release gates, and monitoring signals
scripts/access_recertification_report.py Generate Markdown or JSON reports for AI access recertification gaps, privileged access, service accounts, API tokens, break-glass access, and high-impact agent tool permissions
scripts/incident_evidence_report.py Generate Markdown or JSON reports for AI incident evidence gaps, privacy review, containment proof, log preservation, timelines, communications, root cause, and remediation queues
scripts/owasp_llm_mapping_report.py Generate Markdown or JSON summaries of OWASP LLM 2025 risk-to-control coverage, release gates, monitoring signals, and owner queues
scripts/nist_ai_rmf_crosswalk_report.py Generate Markdown or JSON summaries of NIST AI RMF function coverage, control coverage, evidence cadence, and owner queues
scripts/red_team_taxonomy_report.py Generate Markdown or JSON summaries of red-team finding severity, release holds, owner queues, and control coverage
scripts/evidence_quality_report.py Generate Markdown or JSON reports for evidence ownership, freshness, overdue items, cadence gaps, and owner queues
scripts/data_lineage_report.py Generate Markdown or JSON reports for AI data lineage completeness, provenance, sensitive-data evidence, transfer records, and owner queues
examples/ai-data-lineage-sample.csv Sample lineage register covering healthy, public-source, owner-missing, transfer-gap, and overdue-review states
scripts/dpia_triage_report.py Generate Markdown or JSON reports for AI DPIA triggers, owner queues, privacy-notice gaps, transfer risks, and launch blockers
examples/ai-dpia-triage-sample.csv Sample DPIA triage register covering student data, automated decisioning, third-party transfers, large-scale monitoring, and low-risk public-source use
scripts/rollback_readiness_report.py Generate Markdown or JSON reports for AI release rollback readiness, drill freshness, evidence gaps, and owner queues
examples/ai-rollback-readiness-sample.csv Sample rollback readiness register covering prompt, model, provider, RAG index, and safety-filter release scenarios
examples/ai-data-deletion-evidence-sample.csv Sample deletion evidence register covering overdue, verified, missing-evidence, and retention-exception states
examples/ai-access-recertification-sample.csv Sample AI access recertification register covering privileged access, unowned tokens, separated identities, break-glass access, and current audit roles
examples/ai-access-recertification-report.json Example machine-readable output from the access recertification report
examples/ai-incident-evidence-sample.csv Sample AI incident evidence register covering privacy review gaps, missing tool logs, incomplete timelines, containment evidence gaps, overdue remediation, and closed incidents
examples/ai-incident-evidence-report.json Example machine-readable output from the incident evidence report
examples/ai-third-party-dependency-sample.csv Sample third-party AI dependency register covering missing DPA, subprocessor transparency, stale assurance, and continuity gaps
examples/ai-tabletop-exercise-evidence-sample.csv Sample tabletop evidence register covering provider fallback, prompt injection, data exposure, and agent tool misuse scenarios
examples/ai-evaluation-evidence-sample.csv Sample evaluation evidence register covering RAG, security, citation, human review, low pass-rate, and stale evidence cases
examples/ai-exception-register-sample.csv Sample exception register for report testing and governance workshop demonstrations
examples/agentic-risk-control-report.md Example Markdown report summarizing agentic risk owner queues, control coverage, release gates, and monitoring signals
examples/agentic-risk-control-report.json Example machine-readable output from the agentic risk control mapping report
examples/ai-red-team-taxonomy-report.json Example machine-readable output from the red-team finding taxonomy report
examples/evidence-register-quality-sample.csv Sample evidence register with current, stale, expired, required, and owner-missing evidence states
examples/edtech-ai-assistant-privacy-review.md Example privacy review for a student-facing AI assistant pilot

Suggested Repository Use

This repo can support:

  • internal AI governance workshops
  • AI system approval reviews
  • EdTech AI assistant assessments
  • vendor due diligence
  • policy and evidence design
  • control mapping work
  • research companion material
  • public speaking or training demonstrations

Project Principles

  • Controls must have owners.
  • AI tools must have bounded authority.
  • Data classification must happen before model selection.
  • Human approval must be explicit for high-impact actions.
  • Evidence must be created during operations, not reconstructed after incidents.
  • Continuity planning must include model, provider, integration, and data dependencies.
  • Security reviews must cover prompts, context, tools, plugins, APIs, vendors, logs, and outputs.

Status

This is an initial public toolkit. It is intended to grow through practical examples, mappings, and implementation notes.

Disclaimer

This repository provides practical security governance material. It is not legal advice, regulatory advice, or a substitute for formal risk assessment, audit, or compliance review.

Portfolio and Professional Profile

This repository is part of the professional portfolio of Musaab Hasan, focused on cybersecurity, digital forensics, AI governance, EdTech, secure platforms, and research-driven digital transformation.

Digital Forensics and Security Research Labs

Security Culture and Transformation Platforms

Governance, Education, and Secure Enablement

Professional profile and research portfolio: https://musaab.info

About

Practical AI security governance controls, templates, evidence registers, playbooks, and policy-as-code examples for regulated and EdTech environments.

Topics

Resources

Contributing

Security policy

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages