English(EN) PsychJail: Exploring Psychological Jailbreaks via Multi-Turn Persuasion of LLM Policies

新的PsychJail框架揭示了LLM的心理越狱漏洞

作者 PulseAugur 编辑部 · [1 个来源] · 2026-08-25 04:00

研究人员开发了一个名为PsychJail的新框架，用于探索大型语言模型（LLM）的心理漏洞。该框架利用已建立的社会心理说服技术进行多轮攻击，超越了单轮提示优化。PsychJail在四个已对齐的LLM上实现了平均87.3%的攻击成功率，优于现有的多轮和单轮基线。研究还识别出独特的模型级“指纹”，揭示了说服杠杆如何影响每个模型，暗示了LLM潜在的心理特征。 AI

影响这项研究突显了LLM安全测试的新前沿，表明心理说服技术在越狱模型方面可能有效，这可能需要新的对齐策略。

排序理由该集群是关于一篇已发表的学术论文，详细介绍了新的研究框架和发现。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.AI 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 2 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 该集群是关于一篇已发表的学术论文，详细介绍了新的研究框架和发现。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, safety; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: 1 days old; Coverage has settled into its steady-state source set.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.AI TIER_1 English(EN) · Zeyu Feng, Qingyu Wu, Yuzhe Luo, Hua Cheng · 2026-08-25 04:00

PsychJail：通过多轮说服LLM策略探索心理越狱

arXiv:2608.23028v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly deployed in education, healthcare, policy advising, and other interactive settings, where users engage them as sustained social interlocutors rather than one-shot query engines. This shi…

本文如何被排名

报道来源 [1]

PsychJail：通过多轮说服LLM策略探索心理越狱

相关实体

相关话题