跳到正文
arXiv cs.AI· Xurui Li, Kaisong Song·· 2 小时前AI 评分43

MACE:MCTS 引导的对抗共演化框架,从策略搜索到策略学习强化 LLM 安全对齐

From Strategy Search to Policy Learning: MCTS-Guided Adversarial Co-Evolution for LLM Safety Alignment

AI 导读

一篇 arXiv 论文提出 MACE,用 MCTS 引导攻击者与防御者对抗共演化,将越狱攻击搜索转化为双方策略学习信号。其 GS-MCTS 记录攻击的成功频率与路径深度,AC-TGPO 据此做难度感知的信用分配。跨六个模型—数据集设置,防御方将平均 ASR-LR 从最佳基线的 11.9% 降至 8.8%。

正文 · 原文

View PDF HTML (experimental)

Abstract:Large Language Models (LLMs) increasingly power web-facing applications, including conversational assistants and content services, where open user interaction exposes them to diverse safety threats and adaptive attacks. In particular, jailbreak attackers repeatedly reformulate harmful requests to bypass safety alignment. Although existing studies have concentrated on defending dynamic jailbreaks, they often retain discovered attacks as flat samples and discard information about how those attacks were found. In this paper, we introduce MACE, an MCTS-guided adversarial co-evolution framework that turns search into policy-learning signals for both an attacker and a defender. Its search component, GS-MCTS, explores multi-step compositions of rewriting operators and associates discovered attacks with empirical success frequency and path depth. Its optimization component, AC-TGPO, uses these statistics for difficulty-aware group-relative credit assignment and constructs a temporally asymmetric curriculum from current interactions, pairings between the latest attacker and preceding defender, and attacks that survive a defense update. Thus, MACE connects search structure to both within-round policy optimization and cross-round memory, rather than using MCTS only to generate attacks or treating each round as a flat adversarial dataset. Removing difficulty-aware weighting, asymmetric samples, or persistent hard samples increases defense ASR-LR by 3.7 %, 6.0 %, and 5.0 %, respectively. Across evaluated models, the resulting attacker achieves the highest ASR-LR in all six model--dataset settings, while the defender reduces average ASR-LR from 11.9 % for the best baseline to 8.8 %.
Comments: Accepted by WSDM, to be published, update arXiv, version
Subjects: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
Cite as: arXiv:2511.19218 [cs.CR]
  (or arXiv:2511.19218v3 [cs.CR] for this version)
  https://doi.org/10.48550/arXiv.2511.19218

arXiv-issued DOI via DataCite

Related DOI: https://doi.org/10.1145/3862322.3863664

DOI(s) linking to related resources

Submission history

From: Xurui Li [view email]
[v1] Mon, 24 Nov 2025 15:23:41 UTC (1,042 KB)
[v2] Wed, 26 Nov 2025 15:12:18 UTC (1,041 KB)
[v3] Tue, 17 Nov 2026 04:10:48 UTC (7,622 KB)

来源:arXiv cs.AI · arxiv.org