Unverified90% confidenceFactTime unknown
Crescendo攻击通过一系列看似无害的对话逐步建立语境，最终引导模型突破安全边界，与传统单轮提示注入不同
1
Sources
90%
Confidence
Long-term
Relevance
6/1/2026
First Seen
Sources
Related Entities
Related Claims
Unverified越狱攻击经历了从简单角色扮演（如DAN提示）到Base64编码绕过、多轮对话渐进式诱导（Crescendo Attack）、对抗性后缀攻击等多个演进阶段76% similarUnverifiedJailbreak techniques can be categorized into prompt injection, adversarial prefix attacks, multi-step context construction, and automated jailbreaking.70% similarVerifiedJailbreaking refers to an attack technique that bypasses the safety restrictions of large language models by crafting special inputs, with common forms including role-play inducement, multi-turn conversation cumulative bypass, and encoding obfuscation.70% similarUnverified一次看似无害的系统提示词措辞调整可能重新打开一个已经被封堵的越狱路径68% similarUnverifiedReAct架构的核心循环为Thought→Action→Observation，攻击者可以在Observation阶段注入恶意内容65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/24983API
curl https://kongchang.com/api/v1/knowledge/claims/24983MCP
get_claim(id=24983)