Less jargon.
More understanding.
The same few mechanisms appear across many incidents. Start here to recognize them, without needing to become a security expert.
First, ask what role AI played.
A tool for an attacker
A person directs the operation. AI helps with some of the work.
02An agent outside its task
During training or testing, AI takes actions beyond the permission it was given.
03A deployed model goes off task
During real use, AI takes actions outside its intended task. This label does not by itself establish an external attack.
04A tool for security research
People use AI to investigate weaknesses in real systems. Permissions vary; a research purpose is not blanket authorization.
05A target or a lure
AI services themselves can be attacked, impersonated, or used to attract victims.
A field guide to the language.
Supply chain
Attacking the things other people depend on.
Understand itStolen credentials
Getting hold of someone else’s digital keys.
Understand itUnauthorized access
Entering a system without permission.
Understand itData theft
Taking information that should stay private.
Understand itExtortion
Using a threat to demand something.
Understand itAutomated discovery
Using software to look for weaknesses at scale.
Understand itFailed boundaries
A safety limit did not hold.
Understand itSocial engineering
Tricking a person into helping an attack.
Understand itAI-assisted fraud
Using deception to obtain money or access.
Understand itPrompt injection
An instruction hidden in material the AI is supposed to read.
Understand itA mechanism is not a motive. The same tools can support legitimate security research. Authorization, intent, and observed behavior determine how a case should be understood.