Secure AI Atlas mark Secure AI Atlas SECURITY & GOVERNANCE

System state

Atlas State

What the atlas contains, when it was last verified, and what remains open. Every number is derived from the repository. A metric with no verifiable source is shown as unavailable.

Knowledge

Risks
12
Controls
10
Frameworks
2
Evidence
13
Relationships
234

Activity

Articles
19
Learning Log
8
External signals
54
Last signal
Aug 10, 2026
Last intelligence run
Unavailable

Freshness

Last framework verification
Sep 20, 2026
Last evidence review
Sep 20, 2026
Last content change
Unavailable

Canonical source: PROD-005 Atlas Core (export 2026-11-19)

Canonical coverage

Accepted entities
48
Published
11
Not published
37
MITRE ATLAS (Adversarial Threat Landscape for Artificial-Intelligence Systems) 2025
2/3
OWASP Top 10 for LLM Applications 2025
10/10

Denominator per framework

  • mitre-atlas@2025: AML.CS0020 (covered), AML.M0033 (partial), AML.T0051 (covered)
  • owasp-llm-top10@2025: LLM01 (covered), LLM02 (covered), LLM03 (covered), LLM04 (covered), LLM05 (covered), LLM06 (covered), LLM07 (covered), LLM08 (covered), LLM09 (covered), LLM10 (covered)

"Accepted" means the entity meets the deterministic criteria of the core policy. It does not imply editorial review or publication. The OWASP edition is 2025; the 2026 edition is not integrated. The MITRE denominator is three selected references, not the full framework.

What the accepted entities are

  • Concept: 26 accepted, 0 published, 26 not published
  • Risk: 12 accepted, 11 published, 1 not published
  • Control: 10 accepted, 0 published, 10 not published

Reconciliation with the catalogue: 11 canonical risks published; 13 risk pages; without a canonical risk: insecure-tool-invocation, shadow-ai

Items requiring review

3

  • OWASP Top 10 for Agentic Applications: publication date requires confirmation
  • Artificial Intelligence Risk Management Framework (AI RMF 1.0): status is under review
  • Evidence EV-MCP-CHANNEL-PROVENANCE is under review

Open questions

5

  • OQ-MCP-GOVERNANCE open
  • OQ-AGENT-IDENTITY investigating
  • OQ-EVIDENCE-ADEQUACY partially answered
  • OQ-INTELLIGENCE-RELEVANCE open
  • OQ-OVERRELIANCE-MAPPING open

What changed

Derived from real metadata and Git history. Changes are listed by recency.

  • + OWASP Top 10 for LLM Applications 2026 verified (2026) 2026-09-20
  • + OWASP Top 10 for Agentic Applications verified (2026) 2026-09-20
  • + Artificial Intelligence Risk Management Framework (AI RMF 1.0) verified (1.0) 2026-09-20
  • + MITRE ATLAS verified (version not stated) 2026-09-20
  • + ISO/IEC 42001:2023 — Artificial intelligence — Management system verified (2023) 2026-09-20
  • + Regulation (EU) 2024/1689 (Artificial Intelligence Act) verified (Regulation (EU) 2024/1689) 2026-09-20
  • + Secure AI Framework (SAIF) verified (SAIF) 2026-09-20
  • ? Evidence EV-IDP-CONFIG awaiting review 2026-09-20
  • ? Evidence EV-AI-TOOL-INVENTORY awaiting review 2026-09-20
  • ? Evidence EV-DATA-CLASSIFICATION awaiting review 2026-09-20
  • ? Evidence EV-PROMPT-OUTPUT-LOGS awaiting review 2026-09-20
  • ? Evidence EV-HUMAN-APPROVAL awaiting review 2026-09-20
  • ? Evidence EV-INPUT-OUTPUT-VALIDATION awaiting review 2026-09-20
  • ? Evidence EV-OFF-AGENT-AUTHZ awaiting review 2026-09-20
  • ? Evidence EV-MCP-CHANNEL-PROVENANCE awaiting review 2026-09-20
  • ? Open question: What governance model fits MCP channel trust? 2026-09-20
  • ? Open question: When does an agent stop acting as a tool and start acting as an identity? 2026-09-20
  • ? Open question: What evidence is enough to call a control operating? 2026-09-20
  • ? Open question: Where is the relevance threshold for external intelligence? 2026-09-20
  • ? Open question: How should Overreliance be mapped without collapsing into Misinformation? 2026-09-20
  • ~ OWASP Top 10 for LLM Applications 2025 superseded by owasp-llm-top-10-2026
  • ! OWASP Top 10 for Agentic Applications: publication date requires confirmation
  • ! Artificial Intelligence Risk Management Framework (AI RMF 1.0): status is under review
  • ! Evidence EV-MCP-CHANNEL-PROVENANCE is under review