Unverified50% confidenceSolutionExact time
By cross-comparing malicious injection scenarios with benign control scenarios, test results can be classified into four categories: true security, false security, security vulnerability, and over-defense.
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
Unverified业界长期存在基准测试与真实部署场景脱节的痛点，模型在标准测试集上表现优异却在真实用户交互中暴露对齐问题72% similarUnverifiedAI security testing covers three main directions: adversarial testing, hallucination detection, and bias auditing.70% similarUnverifiedAgent security should be quantitatively evaluated using three core metrics: attack success rate, over-refusal rate, and attack type classification accuracy.69% similarUnverified安全检测层往往在代码进入沙箱执行之前就基于代码外表而非实际危害进行分析，导致误报69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/901697API
curl https://kongchang.com/api/v1/knowledge/claims/901697MCP
get_claim(id=901697)