Unverified50% confidenceBenchmarkExact time
deepset-deberta and fmops-distilbert captured 100% of attacks in the tests, but their false positive rate was as high as 98%.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/13/2026
First Seen
Valid until: 1/11/2027
Sources
Related Entities
Related Claims
UnverifiedIn tests with 629 AgentDojo attacks and 97 benign tool outputs, jailbreak-detector-large achieved a capture rate of 51% with a false positive rate of 2%.71% similarUnverifiedIn tasks with propositional constraints, the proportion of models integrating candidates already verified as invalid is as high as 93%–100%.69% similarUnverifiedIn the DeepSWE test, 99.6% of all input tokens were cache hits.68% similarUnverified一个 90% 成功、10% 出错的工具调用系统几乎无法投入实际使用，因为单次错误会污染后续上下文造成级联失败66% similarUnverifiedFine-tuning GridSFM with just 10 scenarios can reduce cost error to below 2% and bring feasibility detection above 90%; using 1,000 scenarios can reach in-sample accuracy levels.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1022426API
curl https://kongchang.com/api/v1/knowledge/claims/1022426MCP
get_claim(id=1022426)