Google AI Introduces Consistency Training for Safer Language Models Under Sycophantic and Jailbreak Style Prompts
"越狱攻击" 相关文章
AI越会思考，越容易被骗？「思维链劫持」攻击成功率超过90%
AI越会思考，越容易被骗？「思维链劫持」攻击成功率超过90%
Agentic Moderation: Multi-Agent Design for Safer Vision-Language Models
可攻可防，越狱成功率近90%！六大主流模型全中招 | EMNLP'25
可攻可防，越狱成功率近90%！六大主流模型全中招 | EMNLP'25
AI黑化如恶魔附体！LARGO攻心三步，潜意识种子瞬间开花 | NeurIPS 2025
Learning to Detect Unknown Jailbreak Attacks in Large Vision-Language Models
When "Competency" in Reasoning Opens the Door to Vulnerability: Jailbreaking LLMs via Novel Complex Ciphers
Imperceptible Jailbreaking against Large Language Models
Imperceptible Jailbreaking against Large Language Models
SafeBehavior: Simulating Human-Like Multistage Reasoning to Mitigate Jailbreak Attacks in Large Language Models
Bidirectional Intention Inference Enhances LLMs' Defense Against Multi-Turn Jailbreak Attacks
Formalization Driven LLM Prompt Jailbreaking via Reinforcement Learning
从 0 到 1 了解大模型安全，看这篇就够了
AdaptiveGuard: Towards Adaptive Runtime Safety for LLM-Powered Software
Many-shot jailbreaking
Expanding our model safety bug bounty program
28种LLM越狱攻击方法汇总（2025.8）
成果分享｜[WWW'25]安全防护让大语言模型变“傻”了吗？
