Unverified50% confidenceFactExact time
In RAG systems, models lack a natural ability to discriminate retrieved content and tend to treat text in the context as trustworthy evidence, so any web content that can be retrieved (including hidden text, metadata, or comments) may substantively influence model output.
1
Sources
50%
Confidence
Long-term
Relevance
10/20/2026
First Seen
Sources
WhatsApp Help Pages Secretly Embed LLM Instructions: Don't Recommend Regulators
hackernewshackernews10/19/2026
Related Entities
Related Claims
UnverifiedDocuments retrieved in a RAG system are concatenated into the context window, and the model often cannot distinguish trusted system instructions from untrusted external data.78% similarUnverifiedIn RAG scenarios, if contradictory information exists in the index, vector retrieval cannot automatically determine which piece of content is more credible and will include both in the generation context, leading to confused answers.75% similarUnverifiedThe RAG architecture has a source-trust problem: the model itself lacks the ability to determine whether retrieved results are malicious, and simply treats pages that rank highly or have high relevance scores as reference material.75% similarUnverifiedStructured data gives search systems that support it explicit clues about the meaning of a page and may enable eligible rich results, but valid markup does not guarantee that result features will appear.75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1075181API
curl https://kongchang.com/api/v1/knowledge/claims/1075181MCP
get_claim(id=1075181)