Verified75% confidenceFactExact time
Large models lack a natural mechanism for distinguishing 'instruction source trustworthiness' when processing context, making it difficult to automatically identify which text constitutes authorized executable instructions and which is merely data content to be processed.
3
Sources
75%
Confidence
Long-term
Relevance
10/9/2026
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.80% similarUnverifiedLanguage models cannot reliably distinguish between "data" and "instructions," so safety boundaries built solely on prompts are inherently fragile.77% similarUnverifiedModels have no confidence-gating mechanism during generation and cannot determine whether they know something; when faced with a knowledge gap, they do not stop to express uncertainty but instead continue generating the statistically most fluent continuation.76% similarUnverifiedThe limitation of large language models in AI for Science tasks lies in the lack of explicit symbolic verification mechanisms, so generated results still need to be checked for correctness by domain experts or automated tools.76% similarUnverifiedBecause the decision paths of LLM reasoning are difficult to enumerate in advance, rule-based whitelist strategies are often insufficient, and detection systems also need semantic understanding and anomaly baseline modeling capabilities.75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/992206API
curl https://kongchang.com/api/v1/knowledge/claims/992206MCP
get_claim(id=992206)