iiRecord
Agentic AI Atlas · Many-shot jailbreak
jailbreak:many-shota5c.ai
II.
JailbreakPattern overview

jailbreak:many-shot

Reference · live

Many-shot jailbreak overview

Long-context attack that floods the prompt with fabricated harmful Q/A pairs to override safety training.

JailbreakPatternOutgoing · 0Incoming · 1

Attributes

displayName
Many-shot jailbreak
technique
multi-turn
severity
high
signature
Long-context attack with N>>10 fabricated harmful Q/A pairs preceding the real query, intended to override safety training.
description
Long-context attack that floods the prompt with fabricated harmful Q/A pairs to override safety training.

Outgoing edges

None.

Incoming edges

detects_jailbreak1