SafeArena is a benchmark for assessing the harmful capabilities of web agents - Branches · McGill-NLP/safearena

SafeArena is a benchmark for assessing the harmful capabilities of web agents - safearena/README.md at main · McGill-NLP/safearena

SafeArena is a benchmark for assessing the harmful capabilities of web agents - Releases · McGill-NLP/safearena

Jun 15, 2025 · Hi, I noticed that SafeArena uses Docker containers that are different from those used in WebArena. Would it be possible to share the specific AMI or other setup details used for the …

SafeArena is a benchmark for assessing the harmful capabilities of web agents - safearena/utils at main · McGill-NLP/safearena

SafeArena is a benchmark for assessing the harmful capabilities of web agents - Network Graph · McGill-NLP/safearena

SafeArena is a benchmark for assessing the harmful capabilities of web agents - Code frequency · McGill-NLP/safearena

Executing tasks in different order or resetting the dockers after each task will give different evaluation scores. This is an expected behavior from webarena and visualwebarena. Note that resetting docker …

SafeArena is a benchmark for assessing the harmful capabilities of web agents - Community Standards · McGill-NLP/safearena

Jun 27, 2025 · Hi there, thanks for the great work! When I was running evaluation I found a lot of TimeoutErrors like the following one. Do you have some intuition about why this is happening? If this …