Agentic AI Hallucinations (a36bfc3f-68a4-5276-acdf-ef5de75a1c52)
In multi-agent systems, one agent’s hallucinated output can become another’s input. This can cause cascading misinformation, particularly if agents defer to each other’s outputs without validation. Example: Agent A misclassifies a vulnerability, Agent B acts on this and takes inappropriate mitigation actions.
Threat-modeling question: Could hallucinated output from one agent propagate and mislead others in multi-agent systems?
| Cluster A | Galaxy A | Cluster B | Galaxy B | Level |
|---|---|---|---|---|
| Discover LLM Hallucinations (782c346d-9af5-4145-b6c6-b9cccdc2c950) | MITRE ATLAS Attack Pattern | Agentic AI Hallucinations (a36bfc3f-68a4-5276-acdf-ef5de75a1c52) | PLOT4ai | 1 |