Unverified50% confidenceOpinionExact time
LLMs suffer from hallucination issues and may generate seemingly reasonable but actually harmful commands, such as incorrectly expanding the scope of file cleanup to the entire home directory.
1
Sources
50%
Confidence
Long-term
Relevance
8/10/2026
First Seen
Sources
Cursor AI Agent执行rm -rf删除主目录：事故复盘与防护指南
Related Claims
UnverifiedLLM outputs are probabilistic, carry hallucination risks, and may misjudge context and invoke tools that should not be invoked.74% similarUnverifiedLLM在处理实时性需求时极易产生幻觉，输出过时或错误的信息68% similarUnverified当前LLM在代码生成上存在幻觉问题、上下文窗口限制和安全隐患三个系统性弱点67% similarUnverifiedLLMs are prone to failure in scenarios involving complex business logic, rare edge cases, or situations requiring cross-file global understanding.66% similarUnverified基于规则的命令拦截工具面临误报与漏报平衡的难题，因为命令危险性往往依赖上下文，如 rm -rf ./temp 无害而 rm -rf $VAR 在变量为空时是灾难65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/726773API
curl https://kongchang.com/api/v1/knowledge/claims/726773MCP
get_claim(id=726773)