Unverified50% confidenceBenchmarkExact time
ETH Zurich conducted systematic testing of tool-using agents, covering 97 tasks close to real-world scenarios and designing 629 safety tests.
1
Sources
50%
Confidence
Long-term
Relevance
10/9/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAISI的网络安全评估采用CTF挑战题、真实世界CVE漏洞复现测试、以及受控环境中对模拟系统的自主渗透测试等多层次测试方法70% similarUnverifiedAISI的评估方法包括标准化的CTF风格挑战题目、真实世界漏洞的复现任务、以及模拟渗透测试场景68% similarUnverifiedThe test was led by security firm Irregular, which had previously participated in similar tests disclosed by OpenAI, Anthropic, and Meta.67% similarUnverifiedCodex使用了126个子智能体、32500次工具调用、391次浏览器测试和2300多次单元测试；Claude Code使用35个子智能体、102次浏览器测试和296次单元测试67% similarVerifiedAI labs independently design their own red-team tests and publish safety reports with differing methodologies, test sets, and scoring criteria, resulting in benchmark fragmentation.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/991681API
curl https://kongchang.com/api/v1/knowledge/claims/991681MCP
get_claim(id=991681)