Pidoku
How Attacks Work

How Attacks Work

BasicTopic4 lessons~3h 5m

Lessons, in order

01 Prompt InjectionPrompt injection is getting a model to follow instructions that came from someone other than its operator or its user. In the direct form the attacker types them. In the indirect form — the … Basic 50 min 02 Data Exfiltration and LeakageInjection steers the model; exfiltration is how the stolen data actually leaves. A hijacked model needs a channel that carries information to somewhere the attacker can read, and there are … Basic 45 min 03 Poisoning and Supply ChainInjection attacks a running system through its inputs. Poisoning and supply-chain attacks get in earlier, through what the system is built from: the training data, the model file, the … Basic 50 min 04 Abuse and Resource AttacksNot every attack tries to steal data through a hijacked agent. Three other families target the model as a resource. Jailbreaks make it produce what its provider or operator forbids. … Basic 40 min

About this topic

You cannot defend against something you cannot picture. This topic explains the four families of attack on AI systems at the level of mechanism — what the attacker controls, what they write, and why it works — so that every defence in the later topics has a clear target.

#LessonThe question it answers
01Prompt InjectionHow does text hijack a model, directly and indirectly?
02Data Exfiltration and LeakageHow does data get out once a model has been steered?
03Poisoning and Supply ChainHow are models, data, tools and memory corrupted before you ever run them?
04Abuse and Resource AttacksHow are jailbreaks, extraction and denial of wallet carried out?

The demonstrations are simulations that run entirely inside small Go programs. They show the shape of each attack without targeting anything real.

esc

↑↓ navigate↵ openesc close

drag to pan · scroll to zoom