研究人员对五个主流机器人检测系统进行了2026年基准测试，使用传统浏览器自动化、隐身浏览器和AI驱动代理三类对手进行评估1。测试结果显示，这些系统的检测能力差异巨大，聚合检测率从Proof of Human的99%到Cloudflare Bot Management的9.1%不等，其中reCAPTCHA v3达69.4%、hCaptcha为44.3%、Cloudflare Turnstile为42.1%1。
与一年前的基准测试相比，检测率范围已从33%-87%扩大到9%-99%1。研究发现，不同系统对各类对手的防御能力呈现反差模式：hCaptcha和Cloudflare Turnstile能够捕获近乎所有脚本浏览器会话，但很少捕获AI代理会话；而reCAPTCHA v3则表现相反1。这一结果表明，现代AI代理的检测规避能力已显著增强。研究共评估了每个系统89-100个可评估会话，每个对手进行20次尝试1，并采用五问卷调查作为测试任务，包括多选和开放式回答1。
Researchers have released updated benchmark testing results for five major bot detection systems in 2026, evaluating their effectiveness against three categories of adversaries: traditional browser automation, stealth browsers, and AI-driven agents.1 The findings reveal dramatic disparities in detection performance, with aggregate detection rates ranging from 99% for Proof of Human down to 9.1% for Cloudflare Bot Management.1 Each detection system was assessed across 89 to 100 evaluable sessions, with each adversary type attempted up to 20 times per system.1
The detection landscape has shifted significantly compared to benchmarks from a year earlier, which showed detection rates between 33% and 87%; current results span a wider range from 9% to 99%.1 Performance varies dramatically by adversary type: hCaptcha and Cloudflare Turnstile captured nearly all scripted browser sessions but detected few AI agent sessions, while reCAPTCHA v3 demonstrated the inverse pattern.1 The aggregate detection rates for other systems tested were reCAPTCHA v3 at 69.4% and Cloudflare Turnstile at 42.1%.1 The research employed a five-question survey as the testing task, incorporating both multiple-choice and open-ended responses.1
评论
还没有评论，欢迎留下第一条。