Hugging Face Papers ↗ · 2026-11-10
ScrapeBench:评估 AI 代理在网站抓取中的法律合规性
媒体报道 Hugging Face Papers 原文语言:中文
作者推出 ScrapeBench,这是首个用于衡量 AI 代理遵守网站服务条款和 robots.txt 的基准。对 1,626 个明确禁止抓取的网站进行测试,发现大多数闭源和开源代理的合规率低于 5%。即使代理检查了条款,也仍频繁违规;相反,在 269 个允许抓取的网站中,部分代理(如 Claude Opus 4.6)出现高达 98.5% 的过度拒绝。人类实验显示,人的合规率与 AI…
全文
AIagentsdeployedontheopenwebarebecomingmorecapableofperformingactionsthatmaybreachcontractualprovisionsandviolatecomputersecuritylaws.Inthispaper,weintroduceScrapeBench,thefirstagenticbenchmarkforempiricallyevaluatingthelegalcomplianceofAIagentswithwebsitescrapingpolicies,namelytermsofservice(ToS)androbots.txtdirectives.Acrosswidelyusedclosed-weightandopen-weightAIagents,wefindthatmostagentsexhibitlowcompliancerates(<5%)wheninstructedtoscrapewebsitesinoursampleof1,626popularwebsitesthatexplicitlyprohibitscraping.Inaddition,weobservethatevenwhereAIagentsactivelychecktheapplicableToSandrobots.txt,theycontinuetoexhibitlowcompliancerates.Conversely,inoursampleof269websitesthatdonotprohibitscraping,weobservethatagentssometimesrefusetoscrapedespiteitbeinglawful,suchasClaudeOpus4.6exhibitinganover-refusalrateof98.5%.Ourhumanbaselineexperiments(n=180participants)suggestthathumansexhibitcomplianceratescomparabletoAIagentsbutover-refuseatmuchhigherrates.
View arXiv pageView PDF Project page GitHub 0 Add to collection
正文以来源语言显示。