Secure AI Atlas mark Secure AI Atlas SECURITY & GOVERNANCE

Evidence

Evidence Catalogue

The fourth link in the chain. A control is a claim until something proves it operates; each record here names the evidence that does that and the control it belongs to.

EV-AI-TOOL-INVENTORY

verified

Approved AI tool inventory with owners and review dates

A maintained register listing approved AI tools, allowed use cases, accountable owners, data limits and the next review date.

ISO/IEC 42001 Clause 8NIST AI RMF MapEU AI Act inventory

Last reviewed: 2026-09-20

EV-DATA-CLASSIFICATION

verified

Classification policy and approved data inventory

The data classification policy plus an inventory mapping data classes to what may be shared, transformed, summarized, retained or logged in AI workflows.

ISO/IEC 42001 A.7NIST AI RMF MapEU AI Act data governance

Last reviewed: 2026-09-20

EV-HUMAN-APPROVAL

verified

Approval records bound to the executed operation

Approval records that bind the displayed action to the operation actually executed, with reviewer identity, timestamp and outcome.

NIST AI RMF GovernISO/IEC 42001 A.9EU AI Act human oversight

Last reviewed: 2026-09-20

EV-IDP-CONFIG

verified

Identity provider configuration and authentication logs

Exported IdP configuration proving SSO and MFA are enforced for approved AI platforms, together with authentication events showing the control operates.

NIST AI RMF GVISO/IEC 42001 A.5EU AI Act human oversight

Last reviewed: 2026-09-20

EV-INPUT-OUTPUT-VALIDATION

verified

Validation rules and blocked-request logs

The validation and sanitization rules applied to agent inputs and outputs, plus logs of blocked or sanitized requests that demonstrate the control fires.

MITRE ATLAS input/output validationOWASP LLMNIST AI RMF Measure

Last reviewed: 2026-09-20

EV-MCP-CHANNEL-PROVENANCE

under review

MCP channel provenance and integrity records

Records proving the provenance and integrity of MCP channel input, the separation of telemetry and operational channels, and channel provenance alongside tool calls.

OWASP Agentic ApplicationsMITRE ATLASZero Trust

Last reviewed: 2026-09-20

EV-OFF-AGENT-AUTHZ

verified

Off-agent policy decision log and hash-chain verification

Policy decision logs from the authorization gateway, including denied requests, plus hash-chain verification proving the audit log is tamper-evident.

OWASP Agentic ApplicationsNIST AI RMF GovernZero Trust

Last reviewed: 2026-09-20

EV-PROMPT-OUTPUT-LOGS

verified

Retention configuration and prompt/output audit samples

Retention and access configuration for prompt and output logging, with sampled records that show review, monitoring and incident response are supported.

NIST AI RMF MeasureISO/IEC 42001 A.6EU AI Act traceability

Last reviewed: 2026-09-20

How to read the catalogue

Follow the chain from a risk to the control that reduces it, then to the framework that demands it, and finally to the evidence that shows the control exists. Records marked verified have been reviewed; under review records are still being validated.