Unverified50% confidenceFactExact time
Large language models are fundamentally unable to distinguish between system instructions and user input at a structural level, making prompt injection attacks technically impossible to fully eliminate.
1
Sources
50%
Confidence
Long-term
Relevance
10/12/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedBecause large models themselves cannot strictly distinguish between instructions and data, prompt injection is fundamentally very difficult to completely eliminate.77% similarUnverifiedLLM本质上无法从架构层面区分被处理的数据和应当执行的指令，这类攻击极难从模型本身彻底根除75% similarUnverifiedRelying solely on keyword filtering or rule engines is insufficient to defend against prompt injection attacks, because attack payloads often disguise themselves as legitimate instructions at the semantic level, requiring a model with language understanding capabilities to effectively identify them.74% similarUnverifiedLarge language models have no genuine introspective access to their own execution traces and can only reconstruct a plausible-looking narrative based on context and training tendencies74% similarUnverifiedThe moderation layer does not share weights with the main model and theoretically cannot be bypassed simultaneously by the same prompt.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1002883API
curl https://kongchang.com/api/v1/knowledge/claims/1002883MCP
get_claim(id=1002883)