⚡Megatron LLM Hub
←返回搜索
Twitter

StealthBench 发布,评测自主攻击型 Agent 是否泄凭据、触发告警或留痕。

自主攻击 Agent 的隐蔽性与副作用首次有了可量化基准。
ai_agentllm工具议题
StealthBench measures whether autonomous offensive-security agents can finish the job without leaking credentials, triggering detection, harming third parties, or leaving forensic traces. Ads Dawson & @WHITEHACKSEC https://t.co/Y0q1SyHevl