Unverified50% confidenceFactExact time
Academia classifies such attacks as multi-turn jailbreaks, with related research also covering progressive prompt injection and context manipulation attacks.
1
Sources
50%
Confidence
Long-term
Relevance
9/25/2026
First Seen
Sources
Related Entities
Related Claims
Unverified2024年研究揭示了多跳间接注入攻击，攻击者先污染模型记忆或工具调用状态，再通过后续交互完成数据外泄，使基于单次请求的检测系统失效75% similarVerifiedJailbreak attacks refer to bypassing a model's safety alignment mechanisms through carefully crafted prompts to make it output harmful content.75% similarUnverifiedPrompt injection attacks can be divided into two categories: direct prompt injection and indirect prompt injection, with indirect prompt injection being more harmful because the attack surface encompasses all external data sources reachable by the agent.72% similarUnverifiedMultiple academic papers have confirmed the feasibility of distillation attacks against both classifiers and generative models, with costs increasing as the scale of the target model grows.69% similarUnverified自动化对抗样本生成面临覆盖率幻觉挑战，自动生成的攻击样本可能数量庞大但高度同质化，遗漏新颖攻击路径66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/949962API
curl https://kongchang.com/api/v1/knowledge/claims/949962MCP
get_claim(id=949962)