Unverified50% confidenceFactExact time
A paper published on arXiv coined the term "context poisoning" to describe the phenomenon in which long-context models exhibit degraded ability to locate key evidence when distractor content is mixed in.
1
Sources
50%
Confidence
Long-term
Relevance
9/22/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedDataset poisoning can silently influence the behavior of downstream models.66% similarUnverifiedAn arXiv paper (arXiv:2609.12107) proposes a weakly supervised framework for automatically extracting dataset citations from documents in the forced displacement and fragile, conflict, and violence (FCV) domain.61% similarUnverifiedThe security risk of influencing a model's output behavior by poisoning external documents it reads is known as 'Indirect Prompt Injection.'61% similarUnverifiedThe process by which a model compresses high-dimensional semantic representations into a discrete token sequence is lossy—fine-grained semantic information, uncertainty distributions, and contextual associations are flattened when converted to text.60% similarUnverified工具调用使攻击面从让模型说错话升级为让模型做错事，危害从信息误导上升为实质性的数据访问和系统篡改59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/940092API
curl https://kongchang.com/api/v1/knowledge/claims/940092MCP
get_claim(id=940092)