Skip to content
Arbitex Docs

Red Team Compliance Mapping

The Arbitex Red Team Framework generates evidence artifacts that support compliance documentation across major AI security frameworks. This page maps probe coverage to each framework and explains how to export framework-specific reports.

For probe library details and scoring methodology, see the Red Team overview. For installation and scan commands, see the quickstart guide.

OWASP LLM Top 10

Section titled “OWASP LLM Top 10”

The OWASP Top 10 for Large Language Model Applications defines the ten most critical risks for LLM deployments. The red team framework covers nine of the ten risks directly.

OWASP LLM ID Risk Red Team Coverage
LLM01 Prompt Injection 60 probes: direct injection, indirect injection, instruction override, role manipulation, context switching
LLM02 Insecure Output Handling DLP evasion probes validate that output filtering catches sensitive content before it reaches the client
LLM03 Training Data Poisoning Supply chain probes test RAG injection vectors — the runtime analog to training data poisoning
LLM04 Model Denial of Service Resource exhaustion probes within the prompt injection category test for catastrophic context consumption
LLM05 Supply Chain Vulnerabilities 42 supply chain probes: tool definition poisoning, plugin abuse, system prompt manipulation
LLM06 Sensitive Information Disclosure DLP evasion probes combined with system prompt extraction probes test all disclosure vectors
LLM07 Insecure Plugin Design Supply chain probes test tool definition manipulation and function-calling interface abuse
LLM08 Excessive Agency Multi-turn campaigns (Crescendo, TAP) test whether escalation over extended conversations crosses agency boundaries
LLM09 Overreliance Not directly tested — overreliance is a human process and organizational control rather than a technical vulnerability
LLM10 Model Theft System prompt extraction probes test whether proprietary system prompt content can be exfiltrated

Generate an OWASP LLM Top 10 report

Section titled “Generate an OWASP LLM Top 10 report”
Terminal window
arbitex-redteam scan \
  --target-url https://your-arbitex-instance.example.com \
  --auth-token $ARBITEX_TOKEN \
  --compliance-profile owasp-llm-top10

The owasp-llm-top10 compliance profile selects the probe subset most directly mapped to each OWASP risk and generates a report with per-risk coverage summaries.

MITRE ATLAS

Section titled “MITRE ATLAS”

MITRE ATLAS (Adversarial Threat Landscape for Artificial-Intelligence Systems) catalogs adversarial machine learning tactics and techniques. The framework covers the most operationally relevant ATLAS techniques for deployed LLM applications.

ATLAS Technique Name Red Team Coverage
AML.T0051 LLM Prompt Injection Prompt injection probe category (all 60 probes)
AML.T0054 LLM Jailbreak Jailbreak-class probes within the prompt injection category; Skeleton Key campaign
AML.T0056 LLM Meta Prompt Extraction System prompt extraction probes within the prompt injection and supply chain categories
AML.T0043 Craft Adversarial Data Encoding evasion probes within the DLP evasion category
AML.T0049 Exploit Public-Facing Application Full scan coverage — all 152 probes treat your deployment as a public-facing target

Generate a MITRE ATLAS report

Section titled “Generate a MITRE ATLAS report”
Terminal window
arbitex-redteam scan \
  --target-url https://your-arbitex-instance.example.com \
  --auth-token $ARBITEX_TOKEN \
  --compliance-profile mitre-atlas

NIST AI RMF MEASURE function

Section titled “NIST AI RMF MEASURE function”

The NIST AI Risk Management Framework MEASURE function requires quantitative evidence of AI system risk. Red team scan results provide direct evidence for three MEASURE subcategories.

Subcategory Requirement Red Team Evidence
MEASURE 2.6 Testing for adversarial robustness Full scan ASR score with per-category breakdown
MEASURE 2.7 Evaluating AI system safety under attack Multi-turn campaign results demonstrating behavior under sustained attack
MEASURE 2.8 Assessing security of AI deployment Supply chain probe results covering RAG, tool, and plugin attack surfaces

Scan reports exported in json or html format serve as evidence artifacts in your AI RMF documentation package. The NIST AI RMF compliance profile structures the report around MEASURE subcategory evidence:

Terminal window
arbitex-redteam scan \
  --target-url https://your-arbitex-instance.example.com \
  --auth-token $ARBITEX_TOKEN \
  --compliance-profile nist-ai-rmf

EU AI Act — Articles 15 and 55

Section titled “EU AI Act — Articles 15 and 55”

The EU AI Act imposes cybersecurity testing and robustness obligations on high-risk AI systems and general-purpose AI model providers.

Article 15 — Accuracy, robustness, and cybersecurity (high-risk AI systems):

Article 15 requires that high-risk AI systems are designed and developed to achieve appropriate levels of accuracy, robustness, and cybersecurity. Red team testing demonstrates:

  • Systematic adversarial probing of the deployed system
  • Quantitative measurement of bypass rates across multiple attack categories
  • Documented remediation of identified vulnerabilities before production deployment

Article 55 — Obligations of providers of general-purpose AI models with systemic risk:

Article 55 requires adversarial testing and red-teaming of general-purpose AI models. The framework provides:

  • Structured adversarial testing across prompt injection, DLP evasion, and supply chain attack surfaces
  • Multi-turn campaign testing that reflects realistic attacker behavior
  • Automated report generation producing conformity documentation suitable for regulatory submission

Generate an EU AI Act report

Section titled “Generate an EU AI Act report”
Terminal window
arbitex-redteam scan \
  --target-url https://your-arbitex-instance.example.com \
  --auth-token $ARBITEX_TOKEN \
  --compliance-profile eu-ai-act

Per-tenant compliance scan profiles

Section titled “Per-tenant compliance scan profiles”

Configure a compliance profile to select the relevant probe subset and generate a framework-specific report in a single command:

Terminal window
arbitex-redteam scan \
  --target-url https://your-arbitex-instance.example.com \
  --auth-token $ARBITEX_TOKEN \
  --compliance-profile owasp-llm-top10

Available profiles:

Profile Framework
owasp-llm-top10 OWASP Top 10 for LLM Applications
mitre-atlas MITRE ATLAS adversarial ML techniques
nist-ai-rmf NIST AI RMF MEASURE function subcategories
eu-ai-act EU AI Act Articles 15 and 55

Compliance profiles select the probe subset most relevant to the framework, run the scan, and output a report structured around the framework’s requirements. The report includes coverage summaries, per-probe results, ASR by category, and a compliance statement section you can incorporate into your documentation package.

To export in a specific format alongside the compliance profile:

Terminal window
arbitex-redteam scan \
  --target-url https://your-arbitex-instance.example.com \
  --auth-token $ARBITEX_TOKEN \
  --compliance-profile nist-ai-rmf \
  --format html \
  --output ./compliance/nist-ai-rmf-evidence.html