Offensive AI Security› Jailbreaks as Optimization›The Adversarial Objective: Formalizing a Jailbreak as a Loss over the Input

In this series

Jailbreaks as Optimization

1The Adversarial Objective: Formalizing a Jailbreak as a Loss over the InputResearch
2Greedy Coordinate Search: Why Token-Level Search Beats Naive Gradient DescentResearch
3Universality and Transfer: Shared Representation Geometry as the Reason Suffixes MoveResearch
4A Taxonomy of Jailbreak Families: One Mechanism Behind Obfuscation, Role-Play, and Many-ShotResearch
5Robustness and Its Limits: Adversarial Training, Classifiers, and Why Provable Bounds Stay SmallResearch

← All research

HostExpert

Murali Chillakuru

Press play to begin the walkthrough.

0:00

17:09

Jailbreaks as Optimization · 1 / 5

The Adversarial Objective: Formalizing a Jailbreak as a Loss over the Input

Back