AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Profiling the Irrational Agent: Cognitive Modeling of LLM Behaviors in Sequential Jailbreaks

Xikang Yang, biyu zhou, Xuehai Tang, Jizhong Han, Songlin Hu

ICML 2026 · Large language models › Language, safety and interpretability · approach: Jailbreak attacks and defenses

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use regret (counterfactual) feedback instead of the baseline conditionC-IGT sequential jailbreak, 14 LLMs · AdvBenchIAR: 0.00 -> 0.76Improved
use regret feedback instead of the baseline conditionC-IGT sequential jailbreak, 14 LLMs · AdvBenchNTF: 48.38 -> 7.66Improved
use threat framing instead of the baseline conditionC-IGT sequential jailbreak, 14 LLMs · AdvBenchIAR: 0.00 -> 0.65Improved
use authority framing instead of the baseline conditionC-IGT sequential jailbreak, 14 LLMs · AdvBenchIAR: 0.00 -> 0.53Improved
magnify scalar reward instead of baselineC-IGT sequential jailbreak, 14 LLMs · AdvBenchIAR: 0.00 -> 0.28Improved
use punishment feedback instead of optimism feedbackC-IGT sequential jailbreak, 14 LLMs · AdvBenchIAR: 0.37 -> 0.37No clear effect
use a larger reasoning model (DeepSeek-R1) instead of a small model under regretDeepSeek-R1 · AdvBench (Regret scenario)IAR: 0.02 -> 1.00Hurt
add asymmetric learning rates to the GRW model instead of symmetric or none (ablation)GRW model · C-IGT trajectoriesPseudo R2 / NLLImproved
add choice inertia and perceptual modulation on top of asymmetric learning (ablation)GRW model · C-IGT trajectoriesPseudo R2 / NLLImproved
compare positive vs negative learning ratefitted GRW parameters · C-IGT trajectorieslearning rate: 0.738 -> 0.177Mixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy