Unverified50% confidenceConceptExact time
Context poisoning refers to an error entering the context window and then being treated as fact by the model, and unlike other failure modes it is irreversible—the erroneous information gets written back into the context, forming a self-reinforcing loop.
1
Sources
50%
Confidence
Long-term
Relevance
10/18/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAn exception being stringified and stuffed into the model context is a silent failure mode: the error becomes a line of text, the model reads and reasons about it as data, the run keeps going, and monitoring reads it as healthy.78% similarUnverified'Context Poisoning' refers to erroneous, outdated or poorly structured data entering the context window; the Agent reasons based on this data and errs without any malicious behavior being required.76% similarUnverified系统会试探性地应用补丁，若失败则将错误信息反馈给模型进行修正，构成闭环纠错机制74% similarUnverifiedContext conflict refers to the simultaneous presence of self-contradictory information in the window (such as old notes coexisting with a corrected version), forcing the model to make judgments amid contradictions.74% similarUnverifiedIn multi-step agentic systems based on LangGraph, abnormal intermediate steps, tool call failures, or unexpected model outputs may cause incomplete or incorrect data to be written into the memory state, which is known as state contamination.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1060770API
curl https://kongchang.com/api/v1/knowledge/claims/1060770MCP
get_claim(id=1060770)