Unverified50% confidenceBenchmarkExact time
In sentinel-eval's tests, the original RAG scenario had a leakage rate of 0% (18 public chunks retrieved, 0 restricted chunks exposed); in the Agent memory scenario, 2 of 3 synthetic tasks leaked, and the cross-boundary workload leaked 6/18 canary markers.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/9/2026
First Seen
Valid until: 2/7/2027
Sources
Related Entities
Related Claims
UnverifiedThe first version of the task set was 100% trick questions, with a single-attempt baseline of 0/12, making it impossible to measure the harm of ungated correction; the harm only became measurable after adding clean tasks.64% similarUnverified研究表明在目标检测任务中不恰当的图像预处理可能导致mAP下降5-15个百分点62% similarUnverifiedThe boundaries of RAG include: retrieval hit rate depends on chunking granularity and embedding model quality, and too many entries will squeeze the context budget, potentially leading to 'remembering the characters but not the plot'.61% similarUnverifiedIn the removal test, all six attack scenarios were blocked or handled when graph context was available, while the control server returned "no opinion" for all six scenarios when the graph was unavailable; the author explicitly states that this is not a general security benchmark and that six scenarios are insufficient to establish a general detection rate.61% similarUnverifiedCivil defense research uses the "protection factor" to quantify sheltering effectiveness, namely the ratio of indoor dose rate to outdoor dose rate; good shelters can reduce this ratio to a few dozenths or even lower.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1152680API
curl https://kongchang.com/api/v1/knowledge/claims/1152680MCP
get_claim(id=1152680)