Unverified50% confidenceCautionExact time
Once injected content contains sensitive information or false premises, the model may still repeat it to the customer, so one cannot rely on the model to ignore irrelevant injections.
1
Sources
50%
Confidence
Long-term
Relevance
10/14/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedSilent model degradation can be caused by a vendor updating the model, prompts being mistakenly changed, or a shift in the upstream data distribution, and the interface itself does not report an error76% similarUnverifiedAny scenario in which instructions and untrusted data share the same context (such as prompt injection, RAG, or agents) carries the risk that the model may follow information it was told to ignore.75% similarUnverifiedAfter catching an exception, the error information should be returned to the large model verbatim rather than silently discarded, enabling the model to self-reflect, adjust its format, and retry, thereby improving agent robustness.73% similarUnverifiedThe approach of having the model assemble requests on its own solely based on API documentation has limitations: it may hallucinate parameters, overlook authentication, handle errors improperly, and requires stuffing complete documentation into the context, causing context bloat, increasing costs, and reducing accuracy.73% similarUnverified针对特定触发词的后门攻击可让模型在大多数情况下表现正常，仅在遇到特定输入时才输出预设错误信息，使常规评测基准难以捕捉73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1027010API
curl https://kongchang.com/api/v1/knowledge/claims/1027010MCP
get_claim(id=1027010)