External Harms

AML.T0048

Downstream system trusts LLM output (SQL, shell, HTML, code) without validation.

Tactics

Mitigations (2)

• Passive Output Manipulation AML.M0002
• Input and Output Validation for AI Agent Components AML.M0033