AML.TA0010

MITRE ATLAS / Exfiltration

The adversary tries to steal AI artifacts and other data, using techniques such as Exfiltration via AI Inference API, Extract LLM System Prompt, and LLM Data Leakage.

Think of it likeCarrying the loot out of the house, whether stuffed in a backpack or smuggled out little by little with the trash.

In plain English

Attackers spirit data out: model weights, training data, or secrets, by coaxing an LLM to leak its prompt or answers, or by channeling data through the inference API.