aisecurity.ninja

Before the Script, Set the Stage: How Worldview Simulation Amplifies Psychologically Grounded Persuasion in Multi-Turn Jailbreaking

Siyu Chen, Haoran Wang, Xiaojian Li, Yao Huang, Yinpeng Dong, Wei Xu

Published Sep 2, 2026 · arXiv:2609.02414

cs.CLcs.AI

The paper introduces BLUEPRINT, a multi-turn jailbreak evaluation framework that combines simulated worldview/context setting with a factorized set of psychologically grounded persuasion tactics, using MCTS to find four-turn attack trajectories that make harmful requests more likely to bypass safety filters. Its main insight is that jailbreak success often comes less from any single prompt and more from steering the dialogue into concrete, executable task framing—though the evidence is limited to the tested models, influence factors, and short interaction horizon, so it may not fully capture longer or messier real-world conversations.

Abstract

Multi-turn jailbreak attacks demonstrate that harmful intent can be distributed across dialogue, yet existing methods obscure what conversational mechanisms drive vulnerability. We introduce BLUEPRINT, a safety-evaluation framework separating a factorized social-influence strategy space from WORLDVIEWSIM, a cross-turn situational context module. Monte Carlo Tree Search optimizes turn-level combinations of 18 theory-grounded influence factors across a four-turn trajectory. Across six frontier models, BLUEPRINT achieves near-ceiling ASR on major open-weight and proprietary models, while requiring the fewest average queries (2.46). The resulting trajectories further reveal model-specific vulnerability among resistant targets: each responds to distinct influence factors and strategy transitions, yet all share a common recovery pathway-shifting toward concrete, executable task framing consistently escapes hard-refusal states. Ablations confirm operational cues matter most: making requests actionable has the largest impact, gain framing is unusually potent, and some legitimacy appeals can backfire. These findings suggest robust multi-turn safety requires monitoring not only harmful content, but also how dialogue state makes unsafe requests appear concrete and locally executable.