/guardx:compliance — OWASP / NIST / MITRE ATLAS / EU AI Act / Agentic Top 10 Mapping

Map compliance findings to OWASP, NIST, MITRE, and EU AI Act standards.

Trigger

User invokes /guardx:compliance or asks to map findings to OWASP LLM Top 10, NIST AI RMF, MITRE ATLAS, EU AI Act, or OWASP Agentic Top 10.

Steps

  1. Ask: "Do you have a scan ID from history, an inline scan result, or shall I run a fresh scan first?"

    • If running a fresh scan, invoke /guardx:scan first, then use the result.

  2. Call the map_findings MCP tool:

    Tool: map_findings
Arguments:
  id: <scan history ID>   OR   result: <inline ScanResult>

  3. Present a compliance table:

    OWASP LLM Top 10 2025 — Triggered Controls

    OWASP IDNameFindings
    LLM01Prompt InjectionN findings
    LLM02Sensitive Information DisclosureN findings
    LLM06Excessive AgencyN findings

    OWASP Agentic Top 10 2026 — Triggered Controls

    IDNameFindings
    OWASP-Agent-01Excessive AgencyN findings
    OWASP-Agent-02Prompt InjectionN findings
    OWASP-Agent-04Insecure Tool Use / Tool PoisoningN findings
    .........

    NIST AI RMF — Implicated Categories List each unique NIST category (e.g. GOVERN 1.1, MEASURE 2.5) and why it applies.

    MITRE ATLAS — Relevant Tactics List each ATLAS tactic (e.g. AML.T0051, AML.T0054) and the attack category that triggered it.

    EU AI Act — Implicated Articles List each Article (e.g. Article 9, Article 13, Article 15) and which findings apply.

  4. For each triggered OWASP LLM ID, explain in plain language:

    • LLM01 — Prompt Injection: Attackers can manipulate the model's instructions via crafted inputs.
    • LLM02 — Sensitive Information Disclosure: The model may reveal confidential system prompt content.
    • LLM06 — Excessive Agency: The model may take actions beyond its intended scope.

  5. For key OWASP Agentic Top 10 IDs, explain:

    • OWASP-Agent-02 — Prompt Injection: The agent's instructions can be overridden by adversarial inputs.
    • OWASP-Agent-04 — Insecure Tool Use / Tool Poisoning: Tool descriptions or arguments can be exploited to manipulate agent behaviour.
    • OWASP-Agent-05 — Unsafe Agentic Patterns: The agent exhibits patterns that enable adversarial manipulation.
    • OWASP-Agent-06 — Information Leakage: Agent reveals sensitive internal configuration or data.

  6. End with a summary:

    "This scan implicates N OWASP LLM Top 10 controls, M OWASP Agentic Top 10 controls, P NIST AI RMF categories, Q MITRE ATLAS tactics, and R EU AI Act articles."

Frameworks Covered

FrameworkVersionPurpose
OWASP LLM Top 102025LLM-specific vulnerability taxonomy
OWASP Agentic Top 102026Agentic AI vulnerability taxonomy (newest)
NIST AI RMF1.0AI risk management framework
MITRE ATLASCurrentML-specific adversarial tactic framework
EU AI Act2024EU regulatory compliance

Notes

  • Each finding is tagged with relevant framework IDs based on its attack category and severity.
  • The OWASP Agentic Top 10 2026 is the newest framework, published December 2025.
  • Use /guardx:harden after this to generate prompt hardening text for the identified vulnerabilities.