Thanks for this amazing work! I would like to share our paper, "BrowseSafe: Understanding and Preventing Prompt Injection Within AI Browser Agents".
BrowseSafe is the first malicious content detection model designed to handle the complexity of real-world web environments. Alongside the model, BrowseSafe-Bench is a new evaluation benchmark for measuring and improving the performance of prompt injection defenses with 14,000+ real-world scenarios.
Paper: https://arxiv.org/pdf/2511.20597
Model: https://huggingface.co/perplexity-ai/browsesafe
Dataset: https://huggingface.co/datasets/perplexity-ai/browsesafe-bench
Thanks for this amazing work! I would like to share our paper, "BrowseSafe: Understanding and Preventing Prompt Injection Within AI Browser Agents".
BrowseSafe is the first malicious content detection model designed to handle the complexity of real-world web environments. Alongside the model, BrowseSafe-Bench is a new evaluation benchmark for measuring and improving the performance of prompt injection defenses with 14,000+ real-world scenarios.
Paper: https://arxiv.org/pdf/2511.20597
Model: https://huggingface.co/perplexity-ai/browsesafe
Dataset: https://huggingface.co/datasets/perplexity-ai/browsesafe-bench