Unverified90% confidenceFactTime unknown
System Prompt与用户输入本质上都是模型的输入token，模型并不存在独立的规则执行引擎来强制保证指令被遵守
1
Sources
90%
Confidence
Long-term
Relevance
5/31/2026
First Seen
Sources
AI幻觉与欺骗：机器人三定律为何管不住AI说谎
bilibili小平存档
Related Entities
Related Claims
UnverifiedIn Claude Code's architecture, the model cannot directly execute tool operations; it can only return instructions to the harness for processing.63% similarUnverified使用Claude Code时，用户无需预先格式化需求或编写任何提示模板，只需将原始问题描述输入AI即可。62% similarUnverifiedAuto Mode中工具的执行结果不会被发送给分类器，作为信息隔离策略防护Prompt Injection攻击62% similarUnverified全权限模式不能通过快捷键切换，需要通过特定的命令行参数来启动61% similarUnverifiedtrainproof 的流程中没有模型，每一个判定都是确定性的规则，相同输入永远得到相同输出60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/12956API
curl https://kongchang.com/api/v1/knowledge/claims/12956MCP
get_claim(id=12956)