← AI security in plain English

AML.TA0012

MITRE ATLAS / Privilege Escalation

The adversary tries to gain higher-level permissions, using techniques such as LLM Jailbreak, AI Agent Tool Invocation, and Escape to Host.

Think of it likeA guest let into the living room finding the master key and quietly unlocking every other room in the house.

In plain English

Attackers turn limited access into broad control: jailbreaking an LLM past its guardrails, abusing an agent's tools, or breaking out onto the host machine.