# Cloud AI Control-to-Evidence Mapper Canonical URL: https://move78int.com/aws-azure-gcp-ai-governance-control-evidence-crosswalk.html Last reviewed: 16 October 2026 Source basis: ACT evidence prompts v1.1 and the primary references below. This is editorial review, not verified controls. Cloud settings are one part of an AI control record. Use this mapper to document implementation evidence and the organisational decisions around a deployment. The tool does not inspect cloud accounts or verify your entries. An evidence summary is a starting point for a scoped review. ## Use one system and one cloud response set Choose AWS, Azure or Google Cloud and enter a non-sensitive system reference. Each cloud has its own in-memory responses for the same system. Switching cloud does not copy answers. New system assessment clears all three clouds and the reference. Reset current cloud clears only its ten control responses and six supplier responses. Entries are held in the open page until reset or reload. Explicit JSON or text export retains a copy; print includes all records for the selected cloud. Do not enter credentials, personal data or raw evidence. This page loads Google Fonts. See [website privacy information](https://move78int.com/privacy.html). The actual service, feature, region, contract and configuration determine which evidence applies. Review application and supplier behaviour separately from the underlying cloud. ## Evidence status and human review Each record has status, evidence reference, reviewer, review date, action owner, next action and due date. Not applicable also needs a reason. Unknown is the default. Missing, configured-only, incomplete and invalid records remain unresolved. Evidence recorded is self-reported, not independently verified. Reviewer-backed exclusions are recorded claims, not proof of non-applicability. - NOT ASSESSED: no responses recorded for that panel and cloud. - CRITICAL GAP: critical evidence is unknown, missing or incomplete, overriding completed records. - EVIDENCE GAPS: other unresolved records remain. - REVIEW REQUIRED: record fields are complete; a human decision is still required. All-excluded assessments are identified explicitly. Critical labels are conservative ACT design choices, not universal legal requirements. No state approves deployment or establishes contractual enforceability. ## Ten control topics Workflow filters display relevant rows only. The full summary always assesses all ten topics, lists unresolved IDs, suggested roles or recorded owners, next actions and hidden unresolved controls. It never calculates a readiness percentage. ### Business scope and risk ownership ID: cw-intake-ownership. Suggested owner: Business owner. Critical evidence topic: yes. Evidence: Scope, accountable executive and delegated risk owners. - AWS: Resource inventory, ownership tags and cost attribution. - Azure: Resource inventory, ownership and cost attribution. - Google Cloud: Resource inventory, ownership and cost attribution. Next action: Complete Asset 01 and obtain a scoped decision. ### Agent identity and permission scope ID: cw-nhi-identity. Suggested owner: Platform owner. Critical evidence topic: yes. Evidence: Actual identity, allowed/denied operations, credential lifetime and observed revocation behaviour. - AWS: IAM role and resource policy evidence. - Azure: Managed identity or service-principal permission evidence. - Google Cloud: Service account or workload identity permission evidence. Next action: Complete the Asset 02 identity and boundary records. ### Consequential actions and spending limits ID: cw-tx-spend-cap. Suggested owner: Business and platform owners. Critical evidence topic: yes. Evidence: Application-level action limits, concurrency/retry tests and human authorization. - AWS: Application or gateway enforcement; budget signals as supporting evidence. - Azure: Application or gateway enforcement; budget signals as supporting evidence. - Google Cloud: Application or gateway enforcement; feature-specific billing controls as supporting evidence. Next action: Test action limits before expanding authority. ### MCP and skill isolation ID: cw-mcp-sandbox. Suggested owner: AppSec reviewer. Critical evidence topic: yes. Evidence: Server/skill version, host access, egress, tool schemas and capability tests. - AWS: Workload isolation, networking and denied-path tests. - Azure: Workload isolation, networking and denied-path tests. - Google Cloud: Workload isolation, networking and denied-path tests. Next action: Complete Assets 03A and 03B for the actual deployment. ### Untrusted content and tool-output handling ID: cw-prompt-injection. Suggested owner: AppSec reviewer. Critical evidence topic: yes. Evidence: Adversarial tests against actual data sources, permissions and tool outputs. - AWS: Application controls, selected guardrail settings and observed tests. - Azure: Application controls, selected guardrail settings and observed tests. - Google Cloud: Application controls, selected guardrail settings and observed tests. Next action: Record failures and restrict untested consequential actions. ### Personal data, retention and transfer review ID: cw-pii-dlp. Suggested owner: Privacy reviewer. Critical evidence topic: yes. Evidence: Data-flow inventory, purpose, locations, retention schedule, deletion and transfer assessment. - AWS: Application and service-specific logging, storage and access evidence. - Azure: Application and service-specific logging, storage and access evidence. - Google Cloud: Application and service-specific logging, storage and access evidence. Next action: Resolve applicable data-handling requirements with the responsible reviewer. ### Application vendor and provider evidence ID: cw-saas-wrapper. Suggested owner: Procurement owner. Critical evidence topic: yes. Evidence: Contracting chain, model/service features, locations, purposes and agreed terms. - AWS: Actual AWS service terms plus application vendor terms. - Azure: Actual selected service terms, feature settings and application vendor terms. - Google Cloud: Actual Google Cloud service terms plus application vendor terms. Next action: Complete supplier questions and route unresolved clauses to legal. ### Human review and change monitoring ID: cw-hitl-drift. Suggested owner: Process owner. Critical evidence topic: no. Evidence: Defined sample, trace references, reviewers, overrides, model versions and justified thresholds. - AWS: Application traces, review records and selected service logs. - Azure: Application traces, review records and selected service logs. - Google Cloud: Application traces, review records and selected service logs. Next action: Complete Asset 05 and record the action for each material deviation. ### Observed containment and safe recovery ID: cw-kill-switch. Suggested owner: Incident commander. Critical evidence topic: yes. Evidence: Tested procedure, observed stop, downstream work, restart behaviour and evidence preservation. - AWS: Running tasks, restart mechanisms, resource-side permissions and downstream queues. - Azure: Running work, restart mechanisms, resource-side permissions and downstream queues. - Google Cloud: Running work, restart mechanisms, resource-side permissions and downstream queues. Next action: Test Asset 06 for the scoped workload without delaying urgent harm containment. ### Exceptions and executive oversight ID: cw-exceptions-board. Suggested owner: Accountable executive. Critical evidence topic: no. Evidence: Scoped decision, expiry, restrictions, owner, actions and board evidence cutoff. - AWS: Cloud posture evidence linked to a human decision record. - Azure: Cloud posture evidence linked to a human decision record. - Google Cloud: Cloud posture evidence linked to a human decision record. Next action: Complete Asset 07 and reconcile Assets 08A and 08B. ## Six supplier evidence checks These always retain a separate six-check scope and use the same conservative record rules. - **Contracting chain** (supplier-01): Application supplier, cloud/model providers, subprocessors and processing roles. - **Data purposes and training** (supplier-02): Permitted purposes, secondary use and training treatment for the actual service. - **Retention and deletion** (supplier-03): Application logs, provider logs, backups and deletion evidence by purpose. - **Locations and transfers** (supplier-04): Processing/support locations, transfer routes and applicable review. - **Feature-specific monitoring** (supplier-05): Actual abuse-monitoring and safety settings, applicability and provider approval where required. - **Contract rights and incident handling** (supplier-06): Executed relevant terms, change notice, security events, rights and exit arrangements. Suggested responsible role: procurement owner with legal/privacy reviewers. Complete the AI vendor and subprocessor review and the AI supplier contract rider and negotiation record; route unresolved issues before a scoped decision. ## ACT records for follow-up ACT: AI Controls Toolkit is one product with eleven editable files across eight workflow areas. Relevant result links point to the actual named assets in the [toolkit inventory](https://move78int.com/ai-controls-toolkit.html). Purchasing a toolkit does not resolve a control gap. [Enquire about ACT](https://move78int.com/support.html?product=ACT). ## Sources and boundaries This is an evidence crosswalk, not a clause-level ISO mapping, compliance determination, deployment approval or proof of signed or enforceable terms. Provider commands and precise configuration recipes are outside this generic mapper. - [NIST AI RMF Playbook](https://www.nist.gov/itl/ai-risk-management-framework/nist-ai-rmf-playbook). - [AWS ECS task definition deregistration](https://docs.aws.amazon.com/AmazonECS/latest/developerguide/deregister-task-definition-v2.html). - [AWS Budgets](https://docs.aws.amazon.com/cost-management/latest/userguide/budgets-managing-costs.html). - [Microsoft data privacy documentation](https://learn.microsoft.com/en-us/azure/foundry/responsible-ai/openai/data-privacy). - [ICO storage limitation guidance](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/data-protection-principles/a-guide-to-the-data-protection-principles/storage-limitation/). Source links support review; they do not verify a deployment, evidence, contract or legal position.