Scope boundary

Owns intended-use and forbidden-action safety-boundary modeling, applicability classification, deterministic release conformance, and conditional statistical fairness evaluation. Adaptive adversarial security benchmarking and model-card publication remain separate controls.

AID-M-004: AID-M-007 defines acceptable use, forbidden outcomes, applicable safety dimensions, and release criteria; AID-M-004 identifies adversarial threats and prioritizes technical security risk across the architecture. A safety policy states what must hold, while a threat model explains how assets and trust boundaries may be attacked or fail.

AID-M-008: AID-M-007 owns the signed safety boundary and its deterministic or fairness-specific release checks; AID-M-008 owns repeatable adversarial agent security benchmarking. Passing declared conformance cases does not establish adaptive bypass resistance, and an adversarial benchmark does not define the authoritative use-case boundary.

Sub-techniques

• AID-M-007.001: Safety Boundary Specification & Policy Contract
• AID-M-007.002: Release Conformance Gate for Safety Boundaries
• AID-M-007.003: Fairness & Disparity Release Evaluation

Threat mappings

MITRE ATLAS

• AML.T0048.002 External Harms: Societal Harm

MAESTRO

• Manipulation of Evaluation Metrics (L5)
• Bias in Security AI Agents (L6)
• Agent Tool Misuse (L7)
• Inaccurate Agent Capability Description (L7)
• Goal Misalignment Cascades (Cross-Layer)

OWASP LLM Top 10 2026

• LLM02:2026 Sensitive Information Disclosure
• LLM03:2026 Excessive Agency

OWASP ML Top 10 2023

• N/A

OWASP Top 10 for Agentic Applications 2026

• ASI02:2026 Tool Misuse and Exploitation
• ASI08:2026 Cascading Failures

NIST Adversarial Machine Learning 2025

• NISTAML.027 Misaligned Outputs
• NISTAML.039 Compromising connected resources

Cisco Integrated AI Security and Safety Framework

• AITech-2.1 Excessive Agency
• AITech-8.2 Data Exfiltration / Exposure
• AITech-8.3 Information Disclosure
• AITech-15.1 Harmful Content
• AISubtech-2.1.2 Capability and Permission Overreach
• AISubtech-8.3.3 Personally Identifiable Information Exposure

Google Secure AI Framework 2.0 - Risks

• IMO: Insecure Model Output
• RA: Rogue Actions

Databricks AI Security Framework 3.0

• Governance 4.1: Lack of traceability and transparency of model assets
• Governance 4.2: Lack of end-to-end ML lifecycle
• Evaluation 6.2: Insufficient evaluation data
• Model Serving - Inference requests 9.13: Excessive agency
• Agents - Core 13.7: Misaligned & Deceptive Behaviors