AML.T0130
AI Agent Response Biasing
Adversaries may manipulate an AI assistant so that it favors adversary-chosen sources, or content in its responses. This degrades the integrity and trustworthiness of the assistant's responses on topics the user may rely on, such as health, finance, or security, without the user being aware that the advice has been skewed.
- Tactic Impact
- Maturity: Realized
- Generative AI
- Agentic AI
Mitigations
- Generative AI GuardrailsAML.M0020
Guardrails are safety controls placed between users, tools, and generative AI models to evaluate prompts, retrieved context, model outputs, and agent actions before they are accepted, executed, or shown to a user.
Technical - AI
From MITRE ATLAS 2026.10, Copyright 2021-2026 MITRE, Apache License 2.0.