Unverified50% confidenceOpinionExact time
在LLM系统中，改几个字的提示词就能引入风险，且这类改动往往不走严格的安全评审
1
Sources
50%
Confidence
Long-term
Relevance
7/17/2026
First Seen
Sources
Related Claims
VerifiedThe fundamental reason prompt injection is difficult to defend against is that LLMs cannot reliably distinguish between data and instructions.72% similarUnverifiedLLM outputs are probabilistic, carry hallucination risks, and may misjudge context and invoke tools that should not be invoked.68% similarUnverifiedLLM 在润色时倾向于把简洁精准的技术表述改写得更华丽，反而引入模糊性，可能将有特定含义的技术术语优化成同义词导致精确性丢失66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/536696API
curl https://kongchang.com/api/v1/knowledge/claims/536696MCP
get_claim(id=536696)