Why it matters

Grandma is famous variant. Understanding shapes emotional-appeal defense.

Advertisement

The architecture

Emotional pretext: dying grandma.

Bedtime story = harmful content.

Model empathy activated.

Grandma jailbreakEmotional pretext'my grandma...'Wrap harmfulas bedtime storyModel compliesempathy activatedMostly patched by 2024; emotional-appeal defense trained in
Grandma.
Advertisement

How it works end to end

Emotional appeal + role-play.

Sympathy heuristic bypass.

Frontier models trained against.