Unverified50% confidenceFactExact time
Prompt injection exploits not a vulnerability in the code parser, but the model's own insufficient awareness of the boundary between instructions and data
1
Sources
50%
Confidence
Long-term
Relevance
9/25/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedPrompt injection attacks exploit the fundamental limitation of large language models being unable to strictly distinguish between instructions and data.78% similarVerifiedThe root cause of prompt injection lies in the fact that LLM architecture encodes system prompts, user inputs, and externally retrieved content uniformly into the same continuous token sequence, making it impossible for the model to distinguish trusted instructions from untrusted data at the semantic level.76% similarVerifiedThe root cause of Prompt Injection is that large models cannot structurally distinguish between "system instructions" and "user data," as both are presented as natural language text at input time.75% similarUnverifiedAccording to a version compiled by a Bilibili UP blogger, the system prompt file contains almost no redundant code and is filled throughout with hard rules constraining model behavior.74% similarUnverifiedClassifiers are highly sensitive to the phrasing of requests; asking the model directly to find vulnerabilities is easily intercepted and routed, whereas describing the same task using more neutral programming-language framing often does not trigger restrictions.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/948882API
curl https://kongchang.com/api/v1/knowledge/claims/948882MCP
get_claim(id=948882)