Unverified50% confidenceFactExact time
大语言模型以统一的token序列处理系统提示、用户输入与检索到的外部内容，从架构层面缺乏可靠的信任边界隔离机制
1
Sources
50%
Confidence
Long-term
Relevance
7/20/2026
First Seen
Sources
Related Claims
UnverifiedThe root cause of prompt injection lies in the fact that LLM architecture encodes system prompts, user inputs, and externally retrieved content uniformly into the same continuous token sequence, making it impossible for the model to distinguish trusted instructions from untrusted data at the semantic level.74% similarUnverified在大语言模型底层，系统提示词、用户输入和模型输出都是同一个token序列的不同片段，依靠特殊分隔符和训练时植入的行为模式维持边界，而非硬件或操作系统级别的隔离72% similarUnverifiedTONGYI Lingma's underlying layer relies on the code understanding and generation capabilities of the Tongyi Qianwen series of large language models.71% similarUnverifiedLarge language models with strong reasoning capabilities can understand code semantic logic and discover new types of defects that have never appeared in known vulnerability databases.71% similarUnverified大语言模型能基于统计模式识别协议结构特征（如帧头、长度字段、命令码、校验和），本质是对已知协议模式的泛化而非真正理解70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/566645API
curl https://kongchang.com/api/v1/knowledge/claims/566645MCP
get_claim(id=566645)