跳转至

MLASTG ATLAS Coverage Statistics

Auto-generated from docs/ATLAS-Mapping/2-atlas-navigator-layer.json. Generated: 2026-06-26 ⚠️ Do not edit manually — run python tools/generate_coverage_stats.py to regenerate.

This page provides a quantitative summary of MLASTG's coverage of the MITRE ATLAS framework, derived automatically from the Navigator JSON data file.

Aggregate Metrics

Metric Value
Total ATLAS techniques mapped 28
Full coverage (score ≥ 80%) 19 (67%)
Partial coverage (0 < score < 80%) 9 (32%)
No coverage (score = 0) 0 (0%)
Weighted coverage score 83.9%

Coverage by ATLAS Tactic

Tactic Techniques Full Partial None Avg Score
Reconnaissance (TA0001) 2 1 1 0 75%
Initial Access (TA0002) 1 0 1 0 50%
Resource Development (TA0003) 2 2 0 0 100%
Execution (TA0005) 12 8 4 0 83%
Persistence (TA0006) 2 2 0 0 100%
Evasion (TA0008) 1 1 0 0 100%
Collection (TA0010) 7 5 2 0 86%
Impact (TA0040) 1 0 1 0 50%
Total 28 19 9 0 83.9%

Technique-Level Detail

Full Coverage Techniques (Score ≥ 80%)

ATLAS ID Technique Score MLASVS Controls MLASTG Tests
AML.T0000 ML Attack Staging 100% PIPELINE-001 —
AML.T0001 Search for Tainted ML Data 100% DATA-001 —
AML.T0003 Compromise ML Supply Chain 100% SUPPLY-001, SUPPLY-002 —
AML.T0043 ML Model Inversion Attack 100% MODEL-003 —
AML.T0010 Exploit ML Model 100% MODEL-001, MODEL-002 —
AML.T0015 Evade ML Model 100% MODEL-001, MODEL-002 —
AML.T0012 Backdoor ML Model 100% MODEL-004, MODEL-005 —
AML.T0024.002 Extract ML Model 100% MODEL-002, MODEL-003 —
AML.T0014 Invert ML Model 100% MODEL-003 —
AML.T0020 Poison Training Data 100% DATA-001, DATA-002 —
AML.T0035 Erode ML Model Integrity 100% MODEL-004, MODEL-005 —
AML.T0029 ML Model Extraction 100% MODEL-002 —
AML.T0043 Exploit Adversarial ML Vulnerability 100% MODEL-001 —
AML.T0051 LLM Prompt Injection 100% LLM-001, LLM-004 —
AML.T0057 LLM Output Handling 100% LLM-002 —
AML.T0054 LLM Jailbreak 100% LLM-005 —
AML.T0056 ML Model Behavioral Manipulation 100% MODEL-004, INFRA-012, INFRA-018 —
AML.T0058 ML Model Inversion Attack 100% MODEL-003 —
AML.T0059 Backdoor ML Model 100% MODEL-004, DATA-001 —

Partial Coverage Techniques (Score < 80%)

ATLAS ID Technique Score MLASVS Controls Gap
AML.T0002 Obtain ML Model from Public Source 50% SUPPLY-002 Identified in Gap Analysis
AML.T0005 Exploit ML Model for Initial Access 50% MODEL-001 Identified in Gap Analysis
AML.T0018 Poison Training Data 50% DATA-001, DATA-002 Identified in Gap Analysis
AML.T0021 Clean Training Data 50% DATA-002 Identified in Gap Analysis
AML.T0031 Exfiltration via ML Inference 50% LLM-002 Identified in Gap Analysis
AML.T0024.002 Denial of ML Service 50% LLM-010 Identified in Gap Analysis
AML.T0040 Establish Adversarial ML Infrastructure 50% INFRA-001 Identified in Gap Analysis
AML.T0053 LLM Plugin Compromise 50% LLM-006, LLM-007 Identified in Gap Analysis
AML.T0057 ML Model Data Exfiltration 50% MODEL-003 Identified in Gap Analysis

Cross-Framework Alignment

Framework MLASTG Coverage Notes
MITRE ATLAS 83.9% weighted 19/28 techniques full; 9 partial
OWASP LLM Top 10 10/10 Complete coverage
OWASP ML Top 10 10/10 Complete coverage
NIST AI RMF All 4 functions Govern, Map, Measure, Manage
EU AI Act GOV controls mapped High-risk categories addressed
ISO/IEC 42001 Partial AI Management System alignment

Regenerating This Page

# Auto-generate from Navigator JSON
python tools/generate_coverage_stats.py

# Dry-run (print to stdout)
python tools/generate_coverage_stats.py --dry-run

# Build site
mkdocs build

References